Documentation
¶
Overview ¶
The mcp package provides an SDK for writing model context protocol clients and servers.
To get started, create either a Client or Server, and connect it to a peer using a Transport. The diagram below illustrates how this works:
Client Server ⇅ (jsonrpc2) ⇅ ClientSession ⇄ Client Transport ⇄ Server Transport ⇄ ServerSession
A Client is an MCP client, which can be configured with various client capabilities. Clients may be connected to a Server instance using the Client.Connect method.
Similarly, a Server is an MCP server, which can be configured with various server capabilities. Servers may be connected to one or more Client instances using the Server.Connect method, which creates a ServerSession.
A Transport connects a bidirectional Connection of jsonrpc2 messages. In practice, transports in the MCP spec are are either client transports or server transports. For example, the StdioTransport is a server transport that communicates over stdin/stdout, and its counterpart is a CommandTransport that communicates with a subprocess over its stdin/stdout.
Some transports may hide more complicated details, such as an SSEClientTransport, which reads messages via server-sent events on a hanging GET request, and writes them to a POST endpoint. Users of this SDK may define their own custom Transports by implementing the Transport interface.
Example (ProgressMiddleware) ¶
This middleware function adds a progress token to every outgoing request from the client.
package main import ( "context" "sync/atomic" "golang.org/x/tools/internal/mcp" ) var nextProgressToken atomic.Int64 // This middleware function adds a progress token to every outgoing request // from the client. func main() { c := mcp.NewClient("test", "v1", nil) c.AddSendingMiddleware(addProgressToken[*mcp.ClientSession]) _ = c } func addProgressToken[S mcp.Session](h mcp.MethodHandler[S]) mcp.MethodHandler[S] { return func(ctx context.Context, s S, method string, params mcp.Params) (result mcp.Result, err error) { params.GetMeta().ProgressToken = nextProgressToken.Add(1) return h(ctx, s, method, params) } }
Index ¶
- Constants
- Variables
- func NewInMemoryTransports() (*InMemoryTransport, *InMemoryTransport)
- func ResourceNotFoundError(uri string) error
- type Annotations
- type CallToolParams
- type CallToolParamsFor
- type CallToolResult
- type CallToolResultFor
- type CancelledParams
- type Client
- func (c *Client) AddReceivingMiddleware(middleware ...Middleware[*ClientSession])
- func (c *Client) AddRoots(roots ...*Root)
- func (c *Client) AddSendingMiddleware(middleware ...Middleware[*ClientSession])
- func (c *Client) Connect(ctx context.Context, t Transport) (cs *ClientSession, err error)
- func (c *Client) RemoveRoots(uris ...string)
- type ClientCapabilities
- type ClientOptions
- type ClientSession
- func (cs *ClientSession) CallTool(ctx context.Context, params *CallToolParams) (*CallToolResult, error)
- func (cs *ClientSession) Close() error
- func (cs *ClientSession) GetPrompt(ctx context.Context, params *GetPromptParams) (*GetPromptResult, error)
- func (cs *ClientSession) ListPrompts(ctx context.Context, params *ListPromptsParams) (*ListPromptsResult, error)
- func (cs *ClientSession) ListResourceTemplates(ctx context.Context, params *ListResourceTemplatesParams) (*ListResourceTemplatesResult, error)
- func (cs *ClientSession) ListResources(ctx context.Context, params *ListResourcesParams) (*ListResourcesResult, error)
- func (cs *ClientSession) ListTools(ctx context.Context, params *ListToolsParams) (*ListToolsResult, error)
- func (cs *ClientSession) NotifyProgress(ctx context.Context, params *ProgressNotificationParams) error
- func (cs *ClientSession) Ping(ctx context.Context, params *PingParams) error
- func (cs *ClientSession) Prompts(ctx context.Context, params *ListPromptsParams) iter.Seq2[*Prompt, error]
- func (cs *ClientSession) ReadResource(ctx context.Context, params *ReadResourceParams) (*ReadResourceResult, error)
- func (cs *ClientSession) ResourceTemplates(ctx context.Context, params *ListResourceTemplatesParams) iter.Seq2[*ResourceTemplate, error]
- func (cs *ClientSession) Resources(ctx context.Context, params *ListResourcesParams) iter.Seq2[*Resource, error]
- func (cs *ClientSession) SetLevel(ctx context.Context, params *SetLevelParams) error
- func (cs *ClientSession) Tools(ctx context.Context, params *ListToolsParams) iter.Seq2[*Tool, error]
- func (cs *ClientSession) Wait() error
- type CommandTransport
- type Connection
- type Content
- type CreateMessageParams
- type CreateMessageResult
- type GetPromptParams
- type GetPromptResult
- type InMemoryTransport
- type InitializeParams
- type InitializeResult
- type InitializedParams
- type JSONRPCID
- type JSONRPCMessage
- type JSONRPCRequest
- type JSONRPCResponse
- type ListPromptsParams
- type ListPromptsResult
- type ListResourceTemplatesParams
- type ListResourceTemplatesResult
- type ListResourcesParams
- type ListResourcesResult
- type ListRootsParams
- type ListRootsResult
- type ListToolsParams
- type ListToolsResult
- type LoggingHandler
- type LoggingHandlerOptions
- type LoggingLevel
- type LoggingMessageParams
- type LoggingTransport
- type Meta
- type MethodHandler
- type Middleware
- type ModelHint
- type ModelPreferences
- type Params
- type PingParams
- type ProgressNotificationParams
- type Prompt
- type PromptArgument
- type PromptHandler
- type PromptListChangedParams
- type PromptMessage
- type ReadResourceParams
- type ReadResourceResult
- type Resource
- type ResourceContents
- type ResourceHandler
- type ResourceListChangedParams
- type ResourceTemplate
- type Result
- type Role
- type Root
- type RootsListChangedParams
- type SSEClientTransport
- type SSEClientTransportOptions
- type SSEHandler
- type SSEServerTransport
- type SamplingCapabilities
- type SamplingMessage
- type SchemaOption
- type Server
- func (s *Server) AddPrompts(prompts ...*ServerPrompt)
- func (s *Server) AddReceivingMiddleware(middleware ...Middleware[*ServerSession])
- func (s *Server) AddResourceTemplates(templates ...*ServerResourceTemplate)
- func (s *Server) AddResources(resources ...*ServerResource)
- func (s *Server) AddSendingMiddleware(middleware ...Middleware[*ServerSession])
- func (s *Server) AddTools(tools ...*ServerTool)
- func (s *Server) Connect(ctx context.Context, t Transport) (*ServerSession, error)
- func (s *Server) RemovePrompts(names ...string)
- func (s *Server) RemoveResourceTemplates(uriTemplates ...string)
- func (s *Server) RemoveResources(uris ...string)
- func (s *Server) RemoveTools(names ...string)
- func (s *Server) Run(ctx context.Context, t Transport) error
- func (s *Server) Sessions() iter.Seq[*ServerSession]
- type ServerOptions
- type ServerPrompt
- type ServerResource
- type ServerResourceTemplate
- type ServerSession
- func (ss *ServerSession) Close() error
- func (ss *ServerSession) CreateMessage(ctx context.Context, params *CreateMessageParams) (*CreateMessageResult, error)
- func (ss *ServerSession) ListRoots(ctx context.Context, params *ListRootsParams) (*ListRootsResult, error)
- func (ss *ServerSession) LoggingMessage(ctx context.Context, params *LoggingMessageParams) error
- func (ss *ServerSession) NotifyProgress(ctx context.Context, params *ProgressNotificationParams) error
- func (ss *ServerSession) Ping(ctx context.Context, params *PingParams) error
- func (ss *ServerSession) Wait() error
- type ServerTool
- type Session
- type SetLevelParams
- type StdioTransport
- type StreamableClientTransport
- type StreamableClientTransportOptions
- type StreamableHTTPHandler
- type StreamableHTTPOptions
- type StreamableServerTransport
- func (t *StreamableServerTransport) Close() error
- func (s *StreamableServerTransport) Connect(context.Context) (Connection, error)
- func (t *StreamableServerTransport) Read(ctx context.Context) (JSONRPCMessage, error)
- func (t *StreamableServerTransport) ServeHTTP(w http.ResponseWriter, req *http.Request)
- func (t *StreamableServerTransport) Write(ctx context.Context, msg JSONRPCMessage) error
- type Tool
- type ToolAnnotations
- type ToolHandler
- type ToolHandlerFor
- type ToolListChangedParams
- type ToolOption
- type Transport
Examples ¶
Constants ¶
const ( LevelDebug = slog.LevelDebug LevelInfo = slog.LevelInfo LevelNotice = (slog.LevelInfo + slog.LevelWarn) / 2 LevelWarning = slog.LevelWarn LevelError = slog.LevelError LevelCritical = slog.LevelError + 4 LevelAlert = slog.LevelError + 8 LevelEmergency = slog.LevelError + 12 )
Logging levels.
const ( // The error code to return when a resource isn't found. // See https://modelcontextprotocol.io/specification/2025-03-26/server/resources#error-handling // However, the code they chose is in the wrong space // (see https://github.com/modelcontextprotocol/modelcontextprotocol/issues/509). // so we pick a different one, arbitrarily for now (until they fix it). // The immediate problem is that jsonprc2 defines -32002 as "server closing". CodeResourceNotFound = -31002 // The error code if the method exists and was called properly, but the peer does not support it. CodeUnsupportedMethod = -31001 )
Error codes
const DefaultPageSize = 1000
Variables ¶
var ErrConnectionClosed = errors.New("connection closed")
ErrConnectionClosed is returned when sending a message to a connection that is closed or in the process of closing.
Functions ¶
func NewInMemoryTransports ¶ added in v0.34.0
func NewInMemoryTransports() (*InMemoryTransport, *InMemoryTransport)
NewInMemoryTransports returns two InMemoryTransports that connect to each other.
func ResourceNotFoundError ¶ added in v0.34.0
ResourceNotFoundError returns an error indicating that a resource being read could not be found.
Types ¶
type Annotations ¶ added in v0.34.0
type Annotations struct { // Describes who the intended customer of this object or data is. // // It can include multiple entries to indicate content useful for multiple // audiences (e.g., `["user", "assistant"]`). Audience []Role `json:"audience,omitempty"` // Describes how important this data is for operating the server. // // A value of 1 means "most important," and indicates that the data is // effectively required, while 0 means "least important," and indicates that the // data is entirely optional. Priority float64 `json:"priority,omitempty"` }
Optional annotations for the client. The client can use annotations to inform how objects are used or displayed
type CallToolParams ¶ added in v0.34.0
type CallToolParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Arguments any `json:"arguments,omitempty"` Name string `json:"name"` }
func (*CallToolParams) GetMeta ¶ added in v0.34.0
func (x *CallToolParams) GetMeta() *Meta
type CallToolParamsFor ¶ added in v0.35.0
type CallToolParamsFor[In any] struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Arguments In `json:"arguments,omitempty"` Name string `json:"name"` }
func (*CallToolParamsFor[In]) GetMeta ¶ added in v0.35.0
func (x *CallToolParamsFor[In]) GetMeta() *Meta
type CallToolResult ¶ added in v0.34.0
type CallToolResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Content []*Content `json:"content"` // Whether the tool call ended in an error. // // If not set, this is assumed to be false (the call was successful). IsError bool `json:"isError,omitempty"` }
The server's response to a tool call.
Any errors that originate from the tool SHOULD be reported inside the result object, with `isError` set to true, _not_ as an MCP protocol-level error response. Otherwise, the LLM would not be able to see that an error occurred and self-correct.
However, any errors in _finding_ the tool, an error indicating that the server does not support tool calls, or any other exceptional conditions, should be reported as an MCP error response.
func (*CallToolResult) GetMeta ¶ added in v0.34.0
func (x *CallToolResult) GetMeta() *Meta
type CallToolResultFor ¶ added in v0.35.0
type CallToolResultFor[Out any] struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Content []*Content `json:"content"` // Whether the tool call ended in an error. // // If not set, this is assumed to be false (the call was successful). IsError bool `json:"isError,omitempty"` }
func (*CallToolResultFor[Out]) GetMeta ¶ added in v0.35.0
func (x *CallToolResultFor[Out]) GetMeta() *Meta
type CancelledParams ¶ added in v0.34.0
type CancelledParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An optional string describing the reason for the cancellation. This MAY be // logged or presented to the user. Reason string `json:"reason,omitempty"` // The ID of the request to cancel. // // This MUST correspond to the ID of a request previously issued in the same // direction. RequestID any `json:"requestId"` }
func (*CancelledParams) GetMeta ¶ added in v0.34.0
func (x *CancelledParams) GetMeta() *Meta
type Client ¶
type Client struct {
// contains filtered or unexported fields
}
A Client is an MCP client, which may be connected to an MCP server using the Client.Connect method.
func NewClient ¶
func NewClient(name, version string, opts *ClientOptions) *Client
NewClient creates a new Client.
Use Client.Connect to connect it to an MCP server.
If non-nil, the provided options configure the Client.
func (*Client) AddReceivingMiddleware ¶ added in v0.34.0
func (c *Client) AddReceivingMiddleware(middleware ...Middleware[*ClientSession])
AddReceivingMiddleware wraps the current receiving method handler using the provided middleware. Middleware is applied from right to left, so that the first one is executed first.
For example, AddReceivingMiddleware(m1, m2, m3) augments the method handler as m1(m2(m3(handler))).
Receiving middleware is called when a request is received. It is useful for tasks such as authentication, request logging and metrics.
func (*Client) AddRoots ¶ added in v0.34.0
AddRoots adds the given roots to the client, replacing any with the same URIs, and notifies any connected servers.
func (*Client) AddSendingMiddleware ¶ added in v0.34.0
func (c *Client) AddSendingMiddleware(middleware ...Middleware[*ClientSession])
AddSendingMiddleware wraps the current sending method handler using the provided middleware. Middleware is applied from right to left, so that the first one is executed first.
For example, AddSendingMiddleware(m1, m2, m3) augments the method handler as m1(m2(m3(handler))).
Sending middleware is called when a request is sent. It is useful for tasks such as tracing, metrics, and adding progress tokens.
func (*Client) Connect ¶
Connect begins an MCP session by connecting to a server over the given transport, and initializing the session.
Typically, it is the responsibility of the client to close the connection when it is no longer needed. However, if the connection is closed by the server, calls or notifications will return an error wrapping ErrConnectionClosed.
func (*Client) RemoveRoots ¶ added in v0.34.0
RemoveRoots removes the roots with the given URIs, and notifies any connected servers if the list has changed. It is not an error to remove a nonexistent root. TODO: notification
type ClientCapabilities ¶ added in v0.34.0
type ClientCapabilities struct { // Experimental, non-standard capabilities that the client supports. Experimental map[string]struct { } `json:"experimental,omitempty"` // Present if the client supports listing roots. Roots struct { // Whether the client supports notifications for changes to the roots list. ListChanged bool `json:"listChanged,omitempty"` } `json:"roots,omitempty"` // Present if the client supports sampling from an LLM. Sampling *SamplingCapabilities `json:"sampling,omitempty"` }
Capabilities a client may support. Known capabilities are defined here, in this schema, but this is not a closed set: any client can define its own, additional capabilities.
type ClientOptions ¶
type ClientOptions struct { // Handler for sampling. // Called when a server calls CreateMessage. CreateMessageHandler func(context.Context, *ClientSession, *CreateMessageParams) (*CreateMessageResult, error) // Handlers for notifications from the server. ToolListChangedHandler func(context.Context, *ClientSession, *ToolListChangedParams) PromptListChangedHandler func(context.Context, *ClientSession, *PromptListChangedParams) ResourceListChangedHandler func(context.Context, *ClientSession, *ResourceListChangedParams) LoggingMessageHandler func(context.Context, *ClientSession, *LoggingMessageParams) ProgressNotificationHandler func(context.Context, *ClientSession, *ProgressNotificationParams) }
ClientOptions configures the behavior of the client.
type ClientSession ¶ added in v0.34.0
type ClientSession struct {
// contains filtered or unexported fields
}
A ClientSession is a logical connection with an MCP server. Its methods can be used to send requests or notifications to the server. Create a session by calling Client.Connect.
Call ClientSession.Close to close the connection, or await client termination with ServerSession.Wait.
func (*ClientSession) CallTool ¶ added in v0.34.0
func (cs *ClientSession) CallTool(ctx context.Context, params *CallToolParams) (*CallToolResult, error)
CallTool calls the tool with the given name and arguments. The arguments can be any value that marshals into a JSON object.
func (*ClientSession) Close ¶ added in v0.34.0
func (cs *ClientSession) Close() error
Close performs a graceful close of the connection, preventing new requests from being handled, and waiting for ongoing requests to return. Close then terminates the connection.
func (*ClientSession) GetPrompt ¶ added in v0.34.0
func (cs *ClientSession) GetPrompt(ctx context.Context, params *GetPromptParams) (*GetPromptResult, error)
GetPrompt gets a prompt from the server.
func (*ClientSession) ListPrompts ¶ added in v0.34.0
func (cs *ClientSession) ListPrompts(ctx context.Context, params *ListPromptsParams) (*ListPromptsResult, error)
ListPrompts lists prompts that are currently available on the server.
func (*ClientSession) ListResourceTemplates ¶ added in v0.35.0
func (cs *ClientSession) ListResourceTemplates(ctx context.Context, params *ListResourceTemplatesParams) (*ListResourceTemplatesResult, error)
ListResourceTemplates lists the resource templates that are currently available on the server.
func (*ClientSession) ListResources ¶ added in v0.34.0
func (cs *ClientSession) ListResources(ctx context.Context, params *ListResourcesParams) (*ListResourcesResult, error)
ListResources lists the resources that are currently available on the server.
func (*ClientSession) ListTools ¶ added in v0.34.0
func (cs *ClientSession) ListTools(ctx context.Context, params *ListToolsParams) (*ListToolsResult, error)
ListTools lists tools that are currently available on the server.
func (*ClientSession) NotifyProgress ¶ added in v0.35.0
func (cs *ClientSession) NotifyProgress(ctx context.Context, params *ProgressNotificationParams) error
NotifyProgress sends a progress notification from the client to the server associated with this session. This can be used if the client is performing a long-running task that was initiated by the server
func (*ClientSession) Ping ¶ added in v0.34.0
func (cs *ClientSession) Ping(ctx context.Context, params *PingParams) error
Ping makes an MCP "ping" request to the server.
func (*ClientSession) Prompts ¶ added in v0.34.0
func (cs *ClientSession) Prompts(ctx context.Context, params *ListPromptsParams) iter.Seq2[*Prompt, error]
Prompts provides an iterator for all prompts available on the server, automatically fetching pages and managing cursors. The params argument can set the initial cursor. Iteration stops at the first encountered error, which will be yielded.
func (*ClientSession) ReadResource ¶ added in v0.34.0
func (cs *ClientSession) ReadResource(ctx context.Context, params *ReadResourceParams) (*ReadResourceResult, error)
ReadResource ask the server to read a resource and return its contents.
func (*ClientSession) ResourceTemplates ¶ added in v0.35.0
func (cs *ClientSession) ResourceTemplates(ctx context.Context, params *ListResourceTemplatesParams) iter.Seq2[*ResourceTemplate, error]
ResourceTemplates provides an iterator for all resource templates available on the server, automatically fetching pages and managing cursors. The `params` argument can set the initial cursor. Iteration stops at the first encountered error, which will be yielded.
func (*ClientSession) Resources ¶ added in v0.34.0
func (cs *ClientSession) Resources(ctx context.Context, params *ListResourcesParams) iter.Seq2[*Resource, error]
Resources provides an iterator for all resources available on the server, automatically fetching pages and managing cursors. The params argument can set the initial cursor. Iteration stops at the first encountered error, which will be yielded.
func (*ClientSession) SetLevel ¶ added in v0.34.0
func (cs *ClientSession) SetLevel(ctx context.Context, params *SetLevelParams) error
func (*ClientSession) Tools ¶ added in v0.34.0
func (cs *ClientSession) Tools(ctx context.Context, params *ListToolsParams) iter.Seq2[*Tool, error]
Tools provides an iterator for all tools available on the server, automatically fetching pages and managing cursors. The params argument can set the initial cursor. Iteration stops at the first encountered error, which will be yielded.
func (*ClientSession) Wait ¶ added in v0.34.0
func (cs *ClientSession) Wait() error
Wait waits for the connection to be closed by the server. Generally, clients should be responsible for closing the connection.
type CommandTransport ¶
type CommandTransport struct {
// contains filtered or unexported fields
}
A CommandTransport is a Transport that runs a command and communicates with it over stdin/stdout, using newline-delimited JSON.
func NewCommandTransport ¶
func NewCommandTransport(cmd *exec.Cmd) *CommandTransport
NewCommandTransport returns a CommandTransport that runs the given command and communicates with it over stdin/stdout.
The resulting transport takes ownership of the command, starting it during CommandTransport.Connect, and stopping it when the connection is closed.
func (*CommandTransport) Connect ¶
func (t *CommandTransport) Connect(ctx context.Context) (Connection, error)
Connect starts the command, and connects to it over stdin/stdout.
type Connection ¶ added in v0.35.0
type Connection interface { Read(context.Context) (JSONRPCMessage, error) Write(context.Context, JSONRPCMessage) error Close() error // may be called concurrently by both peers }
A Connection is a logical bidirectional JSON-RPC connection.
type Content ¶
type Content struct { Type string `json:"type"` Text string `json:"text,omitempty"` MIMEType string `json:"mimeType,omitempty"` Data []byte `json:"data,omitempty"` Resource *ResourceContents `json:"resource,omitempty"` Annotations *Annotations `json:"annotations,omitempty"` }
Content is the wire format for content. It represents the protocol types TextContent, ImageContent, AudioContent and EmbeddedResource. Use NewTextContent, NewImageContent, NewAudioContent or NewResourceContent to create one.
The Type field must be one of "text", "image", "audio" or "resource". The constructors above populate this field appropriately. Although at most one of Text, Data, and Resource should be non-zero, consumers of Content use the Type field to determine which value to use; values in the other fields are ignored.
func NewAudioContent ¶ added in v0.34.0
NewAudioContent creates a Content with audio data.
func NewImageContent ¶ added in v0.34.0
NewImageContent creates a Content with image data.
func NewResourceContent ¶ added in v0.34.0
func NewResourceContent(resource *ResourceContents) *Content
NewResourceContent creates a Content with an embedded resource.
func NewTextContent ¶ added in v0.34.0
NewTextContent creates a Content with text.
func (*Content) UnmarshalJSON ¶ added in v0.34.0
type CreateMessageParams ¶ added in v0.34.0
type CreateMessageParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // A request to include context from one or more MCP servers (including the // caller), to be attached to the prompt. The client MAY ignore this request. IncludeContext string `json:"includeContext,omitempty"` // The maximum number of tokens to sample, as requested by the server. The // client MAY choose to sample fewer tokens than requested. MaxTokens int64 `json:"maxTokens"` Messages []*SamplingMessage `json:"messages"` // Optional metadata to pass through to the LLM provider. The format of this // metadata is provider-specific. Metadata struct { } `json:"metadata,omitempty"` // The server's preferences for which model to select. The client MAY ignore // these preferences. ModelPreferences *ModelPreferences `json:"modelPreferences,omitempty"` StopSequences []string `json:"stopSequences,omitempty"` // An optional system prompt the server wants to use for sampling. The client // MAY modify or omit this prompt. SystemPrompt string `json:"systemPrompt,omitempty"` Temperature float64 `json:"temperature,omitempty"` }
func (*CreateMessageParams) GetMeta ¶ added in v0.34.0
func (x *CreateMessageParams) GetMeta() *Meta
type CreateMessageResult ¶ added in v0.34.0
type CreateMessageResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Content *Content `json:"content"` // The name of the model that generated the message. Model string `json:"model"` Role Role `json:"role"` // The reason why sampling stopped, if known. StopReason string `json:"stopReason,omitempty"` }
The client's response to a sampling/create_message request from the server. The client should inform the user before returning the sampled message, to allow them to inspect the response (human in the loop) and decide whether to allow the server to see it.
func (*CreateMessageResult) GetMeta ¶ added in v0.34.0
func (x *CreateMessageResult) GetMeta() *Meta
type GetPromptParams ¶ added in v0.34.0
type GetPromptParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // Arguments to use for templating the prompt. Arguments map[string]string `json:"arguments,omitempty"` // The name of the prompt or prompt template. Name string `json:"name"` }
func (*GetPromptParams) GetMeta ¶ added in v0.34.0
func (x *GetPromptParams) GetMeta() *Meta
type GetPromptResult ¶ added in v0.34.0
type GetPromptResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An optional description for the prompt. Description string `json:"description,omitempty"` Messages []*PromptMessage `json:"messages"` }
The server's response to a prompts/get request from the client.
func (*GetPromptResult) GetMeta ¶ added in v0.34.0
func (x *GetPromptResult) GetMeta() *Meta
type InMemoryTransport ¶ added in v0.34.0
type InMemoryTransport struct {
// contains filtered or unexported fields
}
An InMemoryTransport is a Transport that communicates over an in-memory network connection, using newline-delimited JSON.
type InitializeParams ¶ added in v0.34.0
type InitializeParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Capabilities *ClientCapabilities `json:"capabilities"` ClientInfo *implementation `json:"clientInfo"` // The latest version of the Model Context Protocol that the client supports. // The client MAY decide to support older versions as well. ProtocolVersion string `json:"protocolVersion"` }
func (*InitializeParams) GetMeta ¶ added in v0.34.0
func (x *InitializeParams) GetMeta() *Meta
type InitializeResult ¶ added in v0.34.0
type InitializeResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Capabilities *serverCapabilities `json:"capabilities"` // Instructions describing how to use the server and its features. // // This can be used by clients to improve the LLM's understanding of available // tools, resources, etc. It can be thought of like a "hint" to the model. For // example, this information MAY be added to the system prompt. Instructions string `json:"instructions,omitempty"` // The version of the Model Context Protocol that the server wants to use. This // may not match the version that the client requested. If the client cannot // support this version, it MUST disconnect. ProtocolVersion string `json:"protocolVersion"` ServerInfo *implementation `json:"serverInfo"` }
After receiving an initialize request from the client, the server sends this response.
func (*InitializeResult) GetMeta ¶ added in v0.34.0
func (x *InitializeResult) GetMeta() *Meta
type InitializedParams ¶ added in v0.34.0
type InitializedParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*InitializedParams) GetMeta ¶ added in v0.34.0
func (x *InitializedParams) GetMeta() *Meta
type JSONRPCMessage ¶ added in v0.35.0
JSONRPCMessage is a JSON-RPC message.
type JSONRPCRequest ¶ added in v0.35.0
JSONRPCRequest is a JSON-RPC request.
type JSONRPCResponse ¶ added in v0.35.0
JSONRPCResponse is a JSON-RPC response.
type ListPromptsParams ¶ added in v0.34.0
type ListPromptsParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the current pagination position. If provided, // the server should return results starting after this cursor. Cursor string `json:"cursor,omitempty"` }
func (*ListPromptsParams) GetMeta ¶ added in v0.34.0
func (x *ListPromptsParams) GetMeta() *Meta
type ListPromptsResult ¶ added in v0.34.0
type ListPromptsResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the pagination position after the last returned // result. If present, there may be more results available. NextCursor string `json:"nextCursor,omitempty"` Prompts []*Prompt `json:"prompts"` }
The server's response to a prompts/list request from the client.
func (*ListPromptsResult) GetMeta ¶ added in v0.34.0
func (x *ListPromptsResult) GetMeta() *Meta
type ListResourceTemplatesParams ¶ added in v0.35.0
type ListResourceTemplatesParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the current pagination position. If provided, // the server should return results starting after this cursor. Cursor string `json:"cursor,omitempty"` }
func (*ListResourceTemplatesParams) GetMeta ¶ added in v0.35.0
func (x *ListResourceTemplatesParams) GetMeta() *Meta
type ListResourceTemplatesResult ¶ added in v0.35.0
type ListResourceTemplatesResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the pagination position after the last returned // result. If present, there may be more results available. NextCursor string `json:"nextCursor,omitempty"` ResourceTemplates []*ResourceTemplate `json:"resourceTemplates"` }
The server's response to a resources/templates/list request from the client.
func (*ListResourceTemplatesResult) GetMeta ¶ added in v0.35.0
func (x *ListResourceTemplatesResult) GetMeta() *Meta
type ListResourcesParams ¶ added in v0.34.0
type ListResourcesParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the current pagination position. If provided, // the server should return results starting after this cursor. Cursor string `json:"cursor,omitempty"` }
func (*ListResourcesParams) GetMeta ¶ added in v0.34.0
func (x *ListResourcesParams) GetMeta() *Meta
type ListResourcesResult ¶ added in v0.34.0
type ListResourcesResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the pagination position after the last returned // result. If present, there may be more results available. NextCursor string `json:"nextCursor,omitempty"` Resources []*Resource `json:"resources"` }
The server's response to a resources/list request from the client.
func (*ListResourcesResult) GetMeta ¶ added in v0.34.0
func (x *ListResourcesResult) GetMeta() *Meta
type ListRootsParams ¶ added in v0.34.0
type ListRootsParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*ListRootsParams) GetMeta ¶ added in v0.34.0
func (x *ListRootsParams) GetMeta() *Meta
type ListRootsResult ¶ added in v0.34.0
type ListRootsResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Roots []*Root `json:"roots"` }
The client's response to a roots/list request from the server. This result contains an array of Root objects, each representing a root directory or file that the server can operate on.
func (*ListRootsResult) GetMeta ¶ added in v0.34.0
func (x *ListRootsResult) GetMeta() *Meta
type ListToolsParams ¶ added in v0.34.0
type ListToolsParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the current pagination position. If provided, // the server should return results starting after this cursor. Cursor string `json:"cursor,omitempty"` }
func (*ListToolsParams) GetMeta ¶ added in v0.34.0
func (x *ListToolsParams) GetMeta() *Meta
type ListToolsResult ¶ added in v0.34.0
type ListToolsResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An opaque token representing the pagination position after the last returned // result. If present, there may be more results available. NextCursor string `json:"nextCursor,omitempty"` Tools []*Tool `json:"tools"` }
The server's response to a tools/list request from the client.
func (*ListToolsResult) GetMeta ¶ added in v0.34.0
func (x *ListToolsResult) GetMeta() *Meta
type LoggingHandler ¶ added in v0.34.0
type LoggingHandler struct {
// contains filtered or unexported fields
}
A LoggingHandler is a slog.Handler for MCP.
func NewLoggingHandler ¶ added in v0.34.0
func NewLoggingHandler(ss *ServerSession, opts *LoggingHandlerOptions) *LoggingHandler
NewLoggingHandler creates a LoggingHandler that logs to the given ServerSession using a slog.JSONHandler.
func (*LoggingHandler) Enabled ¶ added in v0.34.0
Enabled implements slog.Handler.Enabled by comparing level to the ServerSession's level.
func (*LoggingHandler) Handle ¶ added in v0.34.0
Handle implements slog.Handler.Handle by writing the Record to a JSONHandler, then calling [ServerSession.LoggingMesssage] with the result.
func (*LoggingHandler) WithAttrs ¶ added in v0.34.0
func (h *LoggingHandler) WithAttrs(as []slog.Attr) slog.Handler
WithAttrs implements slog.Handler.WithAttrs.
func (*LoggingHandler) WithGroup ¶ added in v0.34.0
func (h *LoggingHandler) WithGroup(name string) slog.Handler
WithGroup implements slog.Handler.WithGroup.
type LoggingHandlerOptions ¶ added in v0.34.0
type LoggingHandlerOptions struct { // The value for the "logger" field of logging notifications. LoggerName string // Limits the rate at which log messages are sent. // If zero, there is no rate limiting. MinInterval time.Duration }
LoggingHandlerOptions are options for a LoggingHandler.
type LoggingLevel ¶ added in v0.34.0
type LoggingLevel string
The severity of a log message.
These map to syslog message severities, as specified in RFC-5424: https://datatracker.ietf.org/doc/html/rfc5424#section-6.2.1
type LoggingMessageParams ¶ added in v0.34.0
type LoggingMessageParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // The data to be logged, such as a string message or an object. Any JSON // serializable type is allowed here. Data any `json:"data"` // The severity of this log message. Level LoggingLevel `json:"level"` // An optional name of the logger issuing this message. Logger string `json:"logger,omitempty"` }
func (*LoggingMessageParams) GetMeta ¶ added in v0.34.0
func (x *LoggingMessageParams) GetMeta() *Meta
type LoggingTransport ¶ added in v0.34.0
type LoggingTransport struct {
// contains filtered or unexported fields
}
A LoggingTransport is a Transport that delegates to another transport, writing RPC logs to an io.Writer.
func NewLoggingTransport ¶ added in v0.34.0
func NewLoggingTransport(delegate Transport, w io.Writer) *LoggingTransport
NewLoggingTransport creates a new LoggingTransport that delegates to the provided transport, writing RPC logs to the provided io.Writer.
func (*LoggingTransport) Connect ¶ added in v0.34.0
func (t *LoggingTransport) Connect(ctx context.Context) (Connection, error)
Connect connects the underlying transport, returning a Connection that writes logs to the configured destination.
type Meta ¶ added in v0.34.0
type Meta struct { Data map[string]any `json:",omitempty"` // For params, the progress token can be nil, a string or an integer. // It should be nil for results. ProgressToken any `json:"progressToken,omitempty"` }
func (Meta) MarshalJSON ¶ added in v0.34.0
func (*Meta) UnmarshalJSON ¶ added in v0.34.0
type MethodHandler ¶ added in v0.34.0
type MethodHandler[S Session] func(ctx context.Context, _ S, method string, params Params) (result Result, err error)
A MethodHandler handles MCP messages. For methods, exactly one of the return values must be nil. For notifications, both must be nil.
type Middleware ¶ added in v0.34.0
type Middleware[S Session] func(MethodHandler[S]) MethodHandler[S]
Middleware is a function from MethodHandlers to MethodHandlers.
type ModelHint ¶ added in v0.34.0
type ModelHint struct { // A hint for a model name. // // The client SHOULD treat this as a substring of a model name; for example: - // `claude-3-5-sonnet` should match `claude-3-5-sonnet-20241022` - `sonnet` // should match `claude-3-5-sonnet-20241022`, `claude-3-sonnet-20240229`, etc. - // `claude` should match any Claude model // // The client MAY also map the string to a different provider's model name or a // different model family, as long as it fills a similar niche; for example: - // `gemini-1.5-flash` could match `claude-3-haiku-20240307` Name string `json:"name,omitempty"` }
Hints to use for model selection.
Keys not declared here are currently left unspecified by the spec and are up to the client to interpret.
type ModelPreferences ¶ added in v0.34.0
type ModelPreferences struct { // How much to prioritize cost when selecting a model. A value of 0 means cost // is not important, while a value of 1 means cost is the most important factor. CostPriority float64 `json:"costPriority,omitempty"` // Optional hints to use for model selection. // // If multiple hints are specified, the client MUST evaluate them in order (such // that the first match is taken). // // The client SHOULD prioritize these hints over the numeric priorities, but MAY // still use the priorities to select from ambiguous matches. Hints []*ModelHint `json:"hints,omitempty"` // How much to prioritize intelligence and capabilities when selecting a model. // A value of 0 means intelligence is not important, while a value of 1 means // intelligence is the most important factor. IntelligencePriority float64 `json:"intelligencePriority,omitempty"` // How much to prioritize sampling speed (latency) when selecting a model. A // value of 0 means speed is not important, while a value of 1 means speed is // the most important factor. SpeedPriority float64 `json:"speedPriority,omitempty"` }
The server's preferences for model selection, requested of the client during sampling.
Because LLMs can vary along multiple dimensions, choosing the "best" model is rarely straightforward. Different models excel in different areas—some are faster but less capable, others are more capable but more expensive, and so on. This interface allows servers to express their priorities across multiple dimensions to help clients make an appropriate selection for their use case.
These preferences are always advisory. The client MAY ignore them. It is also up to the client to decide how to interpret these preferences and how to balance them against other considerations.
type Params ¶ added in v0.34.0
type Params interface { // Returns a pointer to the params's Meta field. GetMeta() *Meta }
Params is a parameter (input) type for an MCP call or notification.
type PingParams ¶ added in v0.34.0
type PingParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*PingParams) GetMeta ¶ added in v0.34.0
func (x *PingParams) GetMeta() *Meta
type ProgressNotificationParams ¶ added in v0.35.0
type ProgressNotificationParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // An optional message describing the current progress. Message string `json:"message,omitempty"` // The progress thus far. This should increase every time progress is made, even // if the total is unknown. Progress float64 `json:"progress"` // The progress token which was given in the initial request, used to associate // this notification with the request that is proceeding. ProgressToken any `json:"progressToken"` // Total number of items to process (or total progress required), if known. Total float64 `json:"total,omitempty"` }
func (*ProgressNotificationParams) GetMeta ¶ added in v0.35.0
func (x *ProgressNotificationParams) GetMeta() *Meta
type Prompt ¶
type Prompt struct { // A list of arguments to use for templating the prompt. Arguments []*PromptArgument `json:"arguments,omitempty"` // An optional description of what this prompt provides Description string `json:"description,omitempty"` // The name of the prompt or prompt template. Name string `json:"name"` }
A prompt or prompt template that the server offers.
type PromptArgument ¶ added in v0.34.0
type PromptArgument struct { // A human-readable description of the argument. Description string `json:"description,omitempty"` // The name of the argument. Name string `json:"name"` // Whether this argument must be provided. Required bool `json:"required,omitempty"` }
Describes an argument that a prompt can accept.
type PromptHandler ¶
type PromptHandler func(context.Context, *ServerSession, *GetPromptParams) (*GetPromptResult, error)
A PromptHandler handles a call to prompts/get.
type PromptListChangedParams ¶ added in v0.34.0
type PromptListChangedParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*PromptListChangedParams) GetMeta ¶ added in v0.34.0
func (x *PromptListChangedParams) GetMeta() *Meta
type PromptMessage ¶ added in v0.34.0
Describes a message returned as part of a prompt.
This is similar to `SamplingMessage`, but also supports the embedding of resources from the MCP server.
type ReadResourceParams ¶ added in v0.34.0
type ReadResourceParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // The URI of the resource to read. The URI can use any protocol; it is up to // the server how to interpret it. URI string `json:"uri"` }
func (*ReadResourceParams) GetMeta ¶ added in v0.34.0
func (x *ReadResourceParams) GetMeta() *Meta
type ReadResourceResult ¶ added in v0.34.0
type ReadResourceResult struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` Contents []*ResourceContents `json:"contents"` }
The server's response to a resources/read request from the client.
func (*ReadResourceResult) GetMeta ¶ added in v0.34.0
func (x *ReadResourceResult) GetMeta() *Meta
type Resource ¶
type Resource struct { // Optional annotations for the client. Annotations *Annotations `json:"annotations,omitempty"` // A description of what this resource represents. // // This can be used by clients to improve the LLM's understanding of available // resources. It can be thought of like a "hint" to the model. Description string `json:"description,omitempty"` // The MIME type of this resource, if known. MIMEType string `json:"mimeType,omitempty"` // A human-readable name for this resource. // // This can be used by clients to populate UI elements. Name string `json:"name"` // The size of the raw resource content, in bytes (i.e., before base64 encoding // or any tokenization), if known. // // This can be used by Hosts to display file sizes and estimate context window // usage. Size int64 `json:"size,omitempty"` // The URI of this resource. URI string `json:"uri"` }
A known resource that the server is capable of reading.
type ResourceContents ¶ added in v0.34.0
type ResourceContents struct { URI string `json:"uri"` // resource location; must not be empty MIMEType string `json:"mimeType,omitempty"` Text string `json:"text"` Blob []byte `json:"blob,omitempty"` // if nil, then text; else blob }
A ResourceContents is either a TextResourceContents or a BlobResourceContents. Use NewTextResourceContents or [NextBlobResourceContents] to create one.
func NewBlobResourceContents ¶ added in v0.34.0
func NewBlobResourceContents(uri, mimeType string, blob []byte) *ResourceContents
NewBlobResourceContents returns a ResourceContents containing a byte slice.
func NewTextResourceContents ¶ added in v0.34.0
func NewTextResourceContents(uri, mimeType, text string) *ResourceContents
NewTextResourceContents returns a ResourceContents containing text.
func (ResourceContents) MarshalJSON ¶ added in v0.34.0
func (r ResourceContents) MarshalJSON() ([]byte, error)
type ResourceHandler ¶ added in v0.34.0
type ResourceHandler func(context.Context, *ServerSession, *ReadResourceParams) (*ReadResourceResult, error)
A ResourceHandler is a function that reads a resource. It will be called when the client calls ClientSession.ReadResource. If it cannot find the resource, it should return the result of calling ResourceNotFoundError.
type ResourceListChangedParams ¶ added in v0.34.0
type ResourceListChangedParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*ResourceListChangedParams) GetMeta ¶ added in v0.34.0
func (x *ResourceListChangedParams) GetMeta() *Meta
type ResourceTemplate ¶ added in v0.35.0
type ResourceTemplate struct { // Optional annotations for the client. Annotations *Annotations `json:"annotations,omitempty"` // A description of what this template is for. // // This can be used by clients to improve the LLM's understanding of available // resources. It can be thought of like a "hint" to the model. Description string `json:"description,omitempty"` // The MIME type for all resources that match this template. This should only be // included if all resources matching this template have the same type. MIMEType string `json:"mimeType,omitempty"` // A human-readable name for the type of resource this template refers to. // // This can be used by clients to populate UI elements. Name string `json:"name"` // A URI template (according to RFC 6570) that can be used to construct resource // URIs. URITemplate string `json:"uriTemplate"` }
A template description for resources available on the server.
type Result ¶ added in v0.34.0
type Result interface { // Returns a pointer to the result's Meta field. GetMeta() *Meta }
Result is a result of an MCP call.
type Role ¶ added in v0.34.0
type Role string
The sender or recipient of messages and data in a conversation.
type Root ¶ added in v0.34.0
type Root struct { // An optional name for the root. This can be used to provide a human-readable // identifier for the root, which may be useful for display purposes or for // referencing the root in other parts of the application. Name string `json:"name,omitempty"` // The URI identifying the root. This *must* start with file:// for now. This // restriction may be relaxed in future versions of the protocol to allow other // URI schemes. URI string `json:"uri"` }
Represents a root directory or file that the server can operate on.
type RootsListChangedParams ¶ added in v0.34.0
type RootsListChangedParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*RootsListChangedParams) GetMeta ¶ added in v0.34.0
func (x *RootsListChangedParams) GetMeta() *Meta
type SSEClientTransport ¶
type SSEClientTransport struct {
// contains filtered or unexported fields
}
An SSEClientTransport is a Transport that can communicate with an MCP endpoint serving the SSE transport defined by the 2024-11-05 version of the spec.
https://modelcontextprotocol.io/specification/2024-11-05/basic/transports
func NewSSEClientTransport ¶
func NewSSEClientTransport(baseURL string, opts *SSEClientTransportOptions) *SSEClientTransport
NewSSEClientTransport returns a new client transport that connects to the SSE server at the provided URL.
NewSSEClientTransport panics if the given URL is invalid.
func (*SSEClientTransport) Connect ¶
func (c *SSEClientTransport) Connect(ctx context.Context) (Connection, error)
Connect connects through the client endpoint.
type SSEClientTransportOptions ¶ added in v0.35.0
type SSEClientTransportOptions struct { // HTTPClient is the client to use for making HTTP requests. If nil, // http.DefaultClient is used. HTTPClient *http.Client }
SSEClientTransportOptions provides options for the NewSSEClientTransport constructor.
type SSEHandler ¶
type SSEHandler struct {
// contains filtered or unexported fields
}
SSEHandler is an http.Handler that serves SSE-based MCP sessions as defined by the 2024-11-05 version of the MCP spec.
Example ¶
package main import ( "context" "fmt" "log" "net/http" "net/http/httptest" "golang.org/x/tools/internal/mcp" ) type AddParams struct { X, Y int } func Add(ctx context.Context, cc *mcp.ServerSession, params *mcp.CallToolParamsFor[AddParams]) (*mcp.CallToolResultFor[any], error) { return &mcp.CallToolResultFor[any]{ Content: []*mcp.Content{mcp.NewTextContent(fmt.Sprintf("%d", params.Arguments.X+params.Arguments.Y))}, }, nil } func main() { server := mcp.NewServer("adder", "v0.0.1", nil) server.AddTools(mcp.NewServerTool("add", "add two numbers", Add)) handler := mcp.NewSSEHandler(func(*http.Request) *mcp.Server { return server }) httpServer := httptest.NewServer(handler) defer httpServer.Close() ctx := context.Background() transport := mcp.NewSSEClientTransport(httpServer.URL, nil) client := mcp.NewClient("test", "v1.0.0", nil) cs, err := client.Connect(ctx, transport) if err != nil { log.Fatal(err) } defer cs.Close() res, err := cs.CallTool(ctx, &mcp.CallToolParams{ Name: "add", Arguments: map[string]any{"x": 1, "y": 2}, }) if err != nil { log.Fatal(err) } fmt.Println(res.Content[0].Text) }
Output: 3
func NewSSEHandler ¶
func NewSSEHandler(getServer func(request *http.Request) *Server) *SSEHandler
NewSSEHandler returns a new SSEHandler that creates and manages MCP sessions created via incoming HTTP requests.
Sessions are created when the client issues a GET request to the server, which must accept text/event-stream responses (server-sent events). For each such request, a new SSEServerTransport is created with a distinct messages endpoint, and connected to the server returned by getServer. It is up to the user whether getServer returns a distinct Server for each new request, or reuses an existing server.
The SSEHandler also handles requests to the message endpoints, by delegating them to the relevant server transport.
TODO(rfindley): add options.
func (*SSEHandler) ServeHTTP ¶
func (h *SSEHandler) ServeHTTP(w http.ResponseWriter, req *http.Request)
type SSEServerTransport ¶ added in v0.34.0
type SSEServerTransport struct {
// contains filtered or unexported fields
}
A SSEServerTransport is a logical SSE session created through a hanging GET request.
When connected, it returns the following Connection implementation:
- Writes are SSE 'message' events to the GET response.
- Reads are received from POSTs to the session endpoint, via SSEServerTransport.ServeHTTP.
- Close terminates the hanging GET.
func NewSSEServerTransport ¶ added in v0.34.0
func NewSSEServerTransport(endpoint string, w http.ResponseWriter) *SSEServerTransport
NewSSEServerTransport creates a new SSE transport for the given messages endpoint, and hanging GET response.
Use SSEServerTransport.Connect to initiate the flow of messages.
The transport is itself an http.Handler. It is the caller's responsibility to ensure that the resulting transport serves HTTP requests on the given session endpoint.
Most callers should instead use an SSEHandler, which transparently handles the delegation to SSEServerTransports.
func (*SSEServerTransport) Connect ¶ added in v0.34.0
func (t *SSEServerTransport) Connect(context.Context) (Connection, error)
Connect sends the 'endpoint' event to the client. See SSEServerTransport for more details on the Connection implementation.
func (*SSEServerTransport) ServeHTTP ¶ added in v0.34.0
func (t *SSEServerTransport) ServeHTTP(w http.ResponseWriter, req *http.Request)
ServeHTTP handles POST requests to the transport endpoint.
type SamplingCapabilities ¶ added in v0.34.0
type SamplingCapabilities struct { }
Present if the client supports sampling from an LLM.
type SamplingMessage ¶ added in v0.34.0
Describes a message issued to or received from an LLM API.
type SchemaOption ¶
type SchemaOption interface {
// contains filtered or unexported methods
}
A SchemaOption configures a jsonschema.Schema.
func Description ¶
func Description(desc string) SchemaOption
Description sets the provided schema description.
func Enum ¶
func Enum(values ...any) SchemaOption
Enum sets the provided values as the "enum" value of the schema.
func Property ¶
func Property(name string, opts ...SchemaOption) SchemaOption
Property configures the schema for the property of the given name. If there is no such property in the schema, it is created.
func Required ¶
func Required(v bool) SchemaOption
Required sets whether the associated property is required. It is only valid when used in a Property option: using Required outside of Property panics.
func Schema ¶
func Schema(schema *jsonschema.Schema) SchemaOption
Schema overrides the inferred schema with a shallow copy of the given schema.
type Server ¶
type Server struct {
// contains filtered or unexported fields
}
A Server is an instance of an MCP server.
Servers expose server-side MCP features, which can serve one or more MCP sessions by using [Server.Start] or Server.Run.
Example ¶
package main import ( "context" "fmt" "log" "golang.org/x/tools/internal/mcp" ) type SayHiParams struct { Name string `json:"name"` } func SayHi(ctx context.Context, cc *mcp.ServerSession, params *mcp.CallToolParamsFor[SayHiParams]) (*mcp.CallToolResultFor[any], error) { return &mcp.CallToolResultFor[any]{ Content: []*mcp.Content{ mcp.NewTextContent("Hi " + params.Arguments.Name), }, }, nil } func main() { ctx := context.Background() clientTransport, serverTransport := mcp.NewInMemoryTransports() server := mcp.NewServer("greeter", "v0.0.1", nil) server.AddTools(mcp.NewServerTool("greet", "say hi", SayHi)) serverSession, err := server.Connect(ctx, serverTransport) if err != nil { log.Fatal(err) } client := mcp.NewClient("client", "v0.0.1", nil) clientSession, err := client.Connect(ctx, clientTransport) if err != nil { log.Fatal(err) } res, err := clientSession.CallTool(ctx, &mcp.CallToolParams{ Name: "greet", Arguments: map[string]any{"name": "user"}, }) if err != nil { log.Fatal(err) } fmt.Println(res.Content[0].Text) clientSession.Close() serverSession.Wait() } // createSessions creates and connects an in-memory client and server session for testing purposes. func createSessions(ctx context.Context) (*mcp.ClientSession, *mcp.ServerSession, *mcp.Server) { server := mcp.NewServer("server", "v0.0.1", nil) client := mcp.NewClient("client", "v0.0.1", nil) serverTransport, clientTransport := mcp.NewInMemoryTransports() serverSession, err := server.Connect(ctx, serverTransport) if err != nil { log.Fatal(err) } clientSession, err := client.Connect(ctx, clientTransport) if err != nil { log.Fatal(err) } return clientSession, serverSession, server }
Output: Hi user
func NewServer ¶
func NewServer(name, version string, opts *ServerOptions) *Server
NewServer creates a new MCP server. The resulting server has no features: add features using Server.AddTools, Server.AddPrompts and Server.AddResources.
The server can be connected to one or more MCP clients using [Server.Start] or Server.Run.
If non-nil, the provided options is used to configure the server.
func (*Server) AddPrompts ¶
func (s *Server) AddPrompts(prompts ...*ServerPrompt)
AddPrompts adds the given prompts to the server, replacing any with the same names.
func (*Server) AddReceivingMiddleware ¶ added in v0.34.0
func (s *Server) AddReceivingMiddleware(middleware ...Middleware[*ServerSession])
AddReceivingMiddleware wraps the current receiving method handler using the provided middleware. Middleware is applied from right to left, so that the first one is executed first.
For example, AddReceivingMiddleware(m1, m2, m3) augments the method handler as m1(m2(m3(handler))).
Receiving middleware is called when a request is received. It is useful for tasks such as authentication, request logging and metrics.
func (*Server) AddResourceTemplates ¶ added in v0.35.0
func (s *Server) AddResourceTemplates(templates ...*ServerResourceTemplate)
AddResourceTemplates adds the given resource templates to the server. If a resource template with the same URI template already exists, it will be replaced. AddResourceTemplates panics if a URI template is invalid or not absolute (has an empty scheme).
func (*Server) AddResources ¶ added in v0.34.0
func (s *Server) AddResources(resources ...*ServerResource)
AddResources adds the given resources to the server. If a resource with the same URI already exists, it is replaced. AddResources panics if a resource URI is invalid or not absolute (has an empty scheme).
func (*Server) AddSendingMiddleware ¶ added in v0.34.0
func (s *Server) AddSendingMiddleware(middleware ...Middleware[*ServerSession])
AddSendingMiddleware wraps the current sending method handler using the provided middleware. Middleware is applied from right to left, so that the first one is executed first.
For example, AddSendingMiddleware(m1, m2, m3) augments the method handler as m1(m2(m3(handler))).
Sending middleware is called when a request is sent. It is useful for tasks such as tracing, metrics, and adding progress tokens.
func (*Server) AddTools ¶
func (s *Server) AddTools(tools ...*ServerTool)
AddTools adds the given tools to the server, replacing any with the same names. The arguments must not be modified after this call.
AddTools panics if errors are detected.
func (*Server) Connect ¶
Connect connects the MCP server over the given transport and starts handling messages.
It returns a connection object that may be used to terminate the connection (with [Connection.Close]), or await client termination (with [Connection.Wait]).
func (*Server) RemovePrompts ¶ added in v0.34.0
RemovePrompts removes the prompts with the given names. It is not an error to remove a nonexistent prompt.
func (*Server) RemoveResourceTemplates ¶ added in v0.35.0
RemoveResourceTemplates removes the resource templates with the given URI templates. It is not an error to remove a nonexistent resource.
func (*Server) RemoveResources ¶ added in v0.34.0
RemoveResources removes the resources with the given URIs. It is not an error to remove a nonexistent resource.
func (*Server) RemoveTools ¶ added in v0.34.0
RemoveTools removes the tools with the given names. It is not an error to remove a nonexistent tool.
type ServerOptions ¶
type ServerOptions struct { // Optional instructions for connected clients. Instructions string // If non-nil, called when "notifications/initialized" is received. InitializedHandler func(context.Context, *ServerSession, *InitializedParams) // PageSize is the maximum number of items to return in a single page for // list methods (e.g. ListTools). PageSize int // If non-nil, called when "notifications/roots/list_changed" is received. RootsListChangedHandler func(context.Context, *ServerSession, *RootsListChangedParams) // If non-nil, called when "notifications/progress" is received. ProgressNotificationHandler func(context.Context, *ServerSession, *ProgressNotificationParams) }
ServerOptions is used to configure behavior of the server.
type ServerPrompt ¶ added in v0.34.0
type ServerPrompt struct { Prompt *Prompt Handler PromptHandler }
A Prompt is a prompt definition bound to a prompt handler.
type ServerResource ¶ added in v0.34.0
type ServerResource struct { Resource *Resource Handler ResourceHandler }
A ServerResource associates a Resource with its handler.
type ServerResourceTemplate ¶ added in v0.35.0
type ServerResourceTemplate struct { ResourceTemplate *ResourceTemplate Handler ResourceHandler }
A ServerResourceTemplate associates a ResourceTemplate with its handler.
func (*ServerResourceTemplate) Matches ¶ added in v0.35.0
func (sr *ServerResourceTemplate) Matches(uri string) bool
Matches reports whether the receiver's uri template matches the uri. TODO: use "github.com/yosida95/uritemplate/v3"
type ServerSession ¶ added in v0.34.0
type ServerSession struct {
// contains filtered or unexported fields
}
A ServerSession is a logical connection from a single MCP client. Its methods can be used to send requests or notifications to the client. Create a session by calling Server.Connect.
Call ServerSession.Close to close the connection, or await client termination with ServerSession.Wait.
func (*ServerSession) Close ¶ added in v0.34.0
func (ss *ServerSession) Close() error
Close performs a graceful shutdown of the connection, preventing new requests from being handled, and waiting for ongoing requests to return. Close then terminates the connection.
func (*ServerSession) CreateMessage ¶ added in v0.34.0
func (ss *ServerSession) CreateMessage(ctx context.Context, params *CreateMessageParams) (*CreateMessageResult, error)
CreateMessage sends a sampling request to the client.
func (*ServerSession) ListRoots ¶ added in v0.34.0
func (ss *ServerSession) ListRoots(ctx context.Context, params *ListRootsParams) (*ListRootsResult, error)
ListRoots lists the client roots.
func (*ServerSession) LoggingMessage ¶ added in v0.34.0
func (ss *ServerSession) LoggingMessage(ctx context.Context, params *LoggingMessageParams) error
LoggingMessage sends a logging message to the client. The message is not sent if the client has not called SetLevel, or if its level is below that of the last SetLevel.
TODO(jba): rename to Log or LogMessage. A logging message is the thing that is sent to logging.
func (*ServerSession) NotifyProgress ¶ added in v0.35.0
func (ss *ServerSession) NotifyProgress(ctx context.Context, params *ProgressNotificationParams) error
NotifyProgress sends a progress notification from the server to the client associated with this session. This is typically used to report on the status of a long-running request that was initiated by the client.
func (*ServerSession) Ping ¶ added in v0.34.0
func (ss *ServerSession) Ping(ctx context.Context, params *PingParams) error
Ping pings the client.
func (*ServerSession) Wait ¶ added in v0.34.0
func (ss *ServerSession) Wait() error
Wait waits for the connection to be closed by the client.
type ServerTool ¶ added in v0.34.0
type ServerTool struct { Tool *Tool Handler ToolHandler // contains filtered or unexported fields }
A Tool is a tool definition that is bound to a tool handler.
func NewServerTool ¶ added in v0.35.0
func NewServerTool[In, Out any](name, description string, handler ToolHandlerFor[In, Out], opts ...ToolOption) *ServerTool
NewServerTool is a helper to make a tool using reflection on the given type parameters. When the tool is called, CallToolParams.Arguments will be of type In.
If provided, variadic ToolOption values may be used to customize the tool.
The input schema for the tool is extracted from the request type for the handler, and used to unmmarshal and validate requests to the handler. This schema may be customized using the Input option.
type Session ¶ added in v0.34.0
type Session interface { *ClientSession | *ServerSession // contains filtered or unexported methods }
A Session is either a ClientSession or a ServerSession.
type SetLevelParams ¶ added in v0.34.0
type SetLevelParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` // The level of logging that the client wants to receive from the server. The // server should send all logs at this level and higher (i.e., more severe) to // the client as notifications/message. Level LoggingLevel `json:"level"` }
func (*SetLevelParams) GetMeta ¶ added in v0.34.0
func (x *SetLevelParams) GetMeta() *Meta
type StdioTransport ¶ added in v0.35.0
type StdioTransport struct {
// contains filtered or unexported fields
}
A StdioTransport is a Transport that communicates over stdin/stdout using newline-delimited JSON.
func NewStdioTransport ¶ added in v0.35.0
func NewStdioTransport() *StdioTransport
NewStdioTransport constructs a transport that communicates over stdin/stdout.
type StreamableClientTransport ¶ added in v0.35.0
type StreamableClientTransport struct {
// contains filtered or unexported fields
}
A StreamableClientTransport is a Transport that can communicate with an MCP endpoint serving the streamable HTTP transport defined by the 2025-03-26 version of the spec.
TODO(rfindley): support retries and resumption tokens.
func NewStreamableClientTransport ¶ added in v0.35.0
func NewStreamableClientTransport(url string, opts *StreamableClientTransportOptions) *StreamableClientTransport
NewStreamableClientTransport returns a new client transport that connects to the streamable HTTP server at the provided URL.
func (*StreamableClientTransport) Connect ¶ added in v0.35.0
func (t *StreamableClientTransport) Connect(ctx context.Context) (Connection, error)
Connect implements the Transport interface.
The resulting Connection writes messages via POST requests to the transport URL with the Mcp-Session-Id header set, and reads messages from hanging requests.
When closed, the connection issues a DELETE request to terminate the logical session.
type StreamableClientTransportOptions ¶ added in v0.35.0
type StreamableClientTransportOptions struct { // HTTPClient is the client to use for making HTTP requests. If nil, // http.DefaultClient is used. HTTPClient *http.Client }
StreamableClientTransportOptions provides options for the NewStreamableClientTransport constructor.
type StreamableHTTPHandler ¶ added in v0.35.0
type StreamableHTTPHandler struct {
// contains filtered or unexported fields
}
A StreamableHTTPHandler is an http.Handler that serves streamable MCP sessions, as defined by the MCP spec.
func NewStreamableHTTPHandler ¶ added in v0.35.0
func NewStreamableHTTPHandler(getServer func(*http.Request) *Server, opts *StreamableHTTPOptions) *StreamableHTTPHandler
NewStreamableHTTPHandler returns a new StreamableHTTPHandler.
The getServer function is used to create or look up servers for new sessions. It is OK for getServer to return the same server multiple times.
func (*StreamableHTTPHandler) ServeHTTP ¶ added in v0.35.0
func (h *StreamableHTTPHandler) ServeHTTP(w http.ResponseWriter, req *http.Request)
type StreamableHTTPOptions ¶ added in v0.35.0
type StreamableHTTPOptions struct { }
StreamableHTTPOptions is a placeholder options struct for future configuration of the StreamableHTTP handler.
type StreamableServerTransport ¶ added in v0.35.0
type StreamableServerTransport struct {
// contains filtered or unexported fields
}
A StreamableServerTransport implements the Transport interface for a single session.
func NewStreamableServerTransport ¶ added in v0.35.0
func NewStreamableServerTransport(sessionID string) *StreamableServerTransport
NewStreamableServerTransport returns a new StreamableServerTransport with the given session ID.
A StreamableServerTransport implements the server-side of the streamable transport.
TODO(rfindley): consider adding options here, to configure event storage policy.
func (*StreamableServerTransport) Close ¶ added in v0.35.0
func (t *StreamableServerTransport) Close() error
Close implements the Connection interface.
func (*StreamableServerTransport) Connect ¶ added in v0.35.0
func (s *StreamableServerTransport) Connect(context.Context) (Connection, error)
Connect implements the Transport interface.
TODO(rfindley): Connect should return a new object.
func (*StreamableServerTransport) Read ¶ added in v0.35.0
func (t *StreamableServerTransport) Read(ctx context.Context) (JSONRPCMessage, error)
Read implements the Connection interface.
func (*StreamableServerTransport) ServeHTTP ¶ added in v0.35.0
func (t *StreamableServerTransport) ServeHTTP(w http.ResponseWriter, req *http.Request)
ServeHTTP handles a single HTTP request for the session.
func (*StreamableServerTransport) Write ¶ added in v0.35.0
func (t *StreamableServerTransport) Write(ctx context.Context, msg JSONRPCMessage) error
Write implements the Connection interface.
type Tool ¶
type Tool struct { // Optional additional tool information. Annotations *ToolAnnotations `json:"annotations,omitempty"` // A human-readable description of the tool. // // This can be used by clients to improve the LLM's understanding of available // tools. It can be thought of like a "hint" to the model. Description string `json:"description,omitempty"` // A JSON Schema object defining the expected parameters for the tool. InputSchema *jsonschema.Schema `json:"inputSchema"` // The name of the tool. Name string `json:"name"` }
Definition for a tool the client can call.
type ToolAnnotations ¶ added in v0.34.0
type ToolAnnotations struct { // If true, the tool may perform destructive updates to its environment. If // false, the tool performs only additive updates. // // (This property is meaningful only when `readOnlyHint == false`) // // Default: true DestructiveHint bool `json:"destructiveHint,omitempty"` // If true, calling the tool repeatedly with the same arguments will have no // additional effect on the its environment. // // (This property is meaningful only when `readOnlyHint == false`) // // Default: false IdempotentHint bool `json:"idempotentHint,omitempty"` // If true, this tool may interact with an "open world" of external entities. If // false, the tool's domain of interaction is closed. For example, the world of // a web search tool is open, whereas that of a memory tool is not. // // Default: true OpenWorldHint bool `json:"openWorldHint,omitempty"` // If true, the tool does not modify its environment. // // Default: false ReadOnlyHint bool `json:"readOnlyHint,omitempty"` // A human-readable title for the tool. Title string `json:"title,omitempty"` }
Additional properties describing a Tool to clients.
NOTE: all properties in ToolAnnotations are **hints**. They are not guaranteed to provide a faithful description of tool behavior (including descriptive properties like `title`).
Clients should never make tool use decisions based on ToolAnnotations received from untrusted servers.
type ToolHandler ¶
type ToolHandler func(context.Context, *ServerSession, *CallToolParamsFor[map[string]any]) (*CallToolResult, error)
A ToolHandler handles a call to tools/call. [CallToolParams.Arguments] will contain a map[string]any that has been validated against the input schema. Perhaps this should be an alias for ToolHandlerFor[map[string]any, map[string]any].
type ToolHandlerFor ¶ added in v0.35.0
type ToolHandlerFor[In, Out any] func(context.Context, *ServerSession, *CallToolParamsFor[In]) (*CallToolResultFor[Out], error)
A ToolHandlerFor handles a call to tools/call with typed arguments and results.
type ToolListChangedParams ¶ added in v0.34.0
type ToolListChangedParams struct { // This property is reserved by the protocol to allow clients and servers to // attach additional metadata to their responses. Meta Meta `json:"_meta,omitempty"` }
func (*ToolListChangedParams) GetMeta ¶ added in v0.34.0
func (x *ToolListChangedParams) GetMeta() *Meta
type ToolOption ¶
type ToolOption interface {
// contains filtered or unexported methods
}
A ToolOption configures the behavior of a Tool.
func Input ¶
func Input(opts ...SchemaOption) ToolOption
Input applies the provided SchemaOption configuration to the tool's input schema.
type Transport ¶
type Transport interface { // Connect returns the logical JSON-RPC connection.. // // It is called exactly once by [Server.Connect] or [Client.Connect]. Connect(ctx context.Context) (Connection, error) }
A Transport is used to create a bidirectional connection between MCP client and server.
Transports should be used for at most one call to Server.Connect or [Client.Start].
Source Files
¶
Directories
¶
Path | Synopsis |
---|---|
examples
|
|
internal
|
|
!+ | |
!+ | |
Package jsonschema is an implementation of the [JSON Schema specification], a JSON-based format for describing the structure of JSON data.
|
Package jsonschema is an implementation of the [JSON Schema specification], a JSON-based format for describing the structure of JSON data. |