documentation-agent

README

Documentation Agent

A2A agent server that provides Context7-style documentation capabilities for your agents

A enterprise-ready Agent-to-Agent (A2A) server that provides AI-powered capabilities through a standardized protocol.

Quick Start

The generated binary is a CLI. start boots the A2A server; --help and --version work as you'd expect.

# Run the agent
go run . start

# Or build and invoke the CLI directly
task build
./bin/documentation-agent --help
./bin/documentation-agent --version
./bin/documentation-agent start

# Or with Docker
docker build -t documentation-agent .
docker run -p 8080:8080 documentation-agent

CLI

Command	Description
documentation-agent start	Start the A2A server (blocks until SIGINT/SIGTERM)
documentation-agent --help	Show top-level help (and per-subcommand with <cmd> --help)
documentation-agent --version	Print the embedded version and exit

Quick Install

Add this agent to your Inference Gateway CLI:

infer agents add documentation-agent http://localhost:8080 \
  --oci ghcr.io/inference-gateway/documentation-agent:latest \
  --run

Features

• ✅ A2A protocol compliant
• ✅ AI-powered capabilities
• ✅ Streaming support
• ✅ State transition history
• ✅ Enterprise-ready
• ✅ Minimal dependencies

Endpoints

• GET /.well-known/agent-card.json - Agent metadata and capabilities
• GET /health - Health check endpoint
• POST /a2a - A2A protocol endpoint

Available Tools

Tool	Description	Parameters
Read	Read a file from disk. Returns its contents, optionally sliced by line offset/limit. Use this to load SKILL.md bodies on demand.	file_path, offset, limit
resolve_library_id	Resolves library name to Context7-compatible library ID and returns matching libraries	libraryName, query
get_library_docs	Fetches up-to-date documentation for a library using Context7-compatible library ID	libraryId, query

Skills (loaded into the system prompt)

Skill	Description	Source
library-documentation-lookup	Use this when you need up-to-date documentation for a third-party library or framework before writing code against it. First resolves the library name to a Context7-compatible ID via resolve_library_id when the caller does not already know it (format '/org/project' or '/org/project/version'), then fetches focused, topic-scoped documentation via get_library_docs. Good for filling in unknowns about specific APIs, hooks, configuration options, or version-specific behavior.	bare scaffold (skills/library-documentation-lookup.md)

Configuration

Configure the agent via environment variables:

Custom Configuration

The following custom configuration variables are available. Defaults are derived from spec.config.* in agent.yaml; the env vars below override them at runtime.

Category	Variable	Default
Tools	TOOLS_READ_ENABLED	true

Environment Variables

Category	Variable	Description	Default
Server	A2A_PORT	Server port	8080
Server	A2A_DEBUG	Enable debug mode	false
Server	A2A_AGENT_URL	Agent URL for internal references	http://localhost:8080
Server	A2A_STREAMING_STATUS_UPDATE_INTERVAL	Streaming status update frequency	1s
Server	A2A_SERVER_READ_TIMEOUT	HTTP server read timeout	120s
Server	A2A_SERVER_WRITE_TIMEOUT	HTTP server write timeout	120s
Server	A2A_SERVER_IDLE_TIMEOUT	HTTP server idle timeout	120s
Server	A2A_SERVER_DISABLE_HEALTHCHECK_LOG	Disable logging for health check requests	true
Agent Metadata	A2A_AGENT_CARD_FILE_PATH	Path to agent card JSON file	.well-known/agent-card.json
LLM Client	A2A_AGENT_CLIENT_PROVIDER	LLM provider (openai, anthropic, azure, ollama, deepseek)	``
LLM Client	A2A_AGENT_CLIENT_MODEL	Model to use	``
LLM Client	A2A_AGENT_CLIENT_API_KEY	API key for LLM provider	-
LLM Client	A2A_AGENT_CLIENT_BASE_URL	Custom LLM API endpoint	-
LLM Client	A2A_AGENT_CLIENT_TIMEOUT	Timeout for LLM requests	30s
LLM Client	A2A_AGENT_CLIENT_MAX_RETRIES	Maximum retries for LLM requests	3
LLM Client	A2A_AGENT_CLIENT_MAX_CHAT_COMPLETION_ITERATIONS	Max chat completion rounds	10
LLM Client	A2A_AGENT_CLIENT_MAX_TOKENS	Maximum tokens for LLM responses	4096
LLM Client	A2A_AGENT_CLIENT_TEMPERATURE	Controls randomness of LLM output	0.7
Capabilities	A2A_CAPABILITIES_STREAMING	Enable streaming responses	true
Capabilities	A2A_CAPABILITIES_PUSH_NOTIFICATIONS	Enable push notifications	false
Capabilities	A2A_CAPABILITIES_STATE_TRANSITION_HISTORY	Track state transitions	true
Task Management	A2A_TASK_RETENTION_MAX_COMPLETED_TASKS	Max completed tasks to keep (0 = unlimited)	100
Task Management	A2A_TASK_RETENTION_MAX_FAILED_TASKS	Max failed tasks to keep (0 = unlimited)	50
Task Management	A2A_TASK_RETENTION_CLEANUP_INTERVAL	Cleanup frequency (0 = manual only)	5m
Storage	A2A_QUEUE_PROVIDER	Storage backend (memory or redis)	memory
Storage	A2A_QUEUE_URL	Redis connection URL (when using Redis)	-
Storage	A2A_QUEUE_MAX_SIZE	Maximum queue size	100
Storage	A2A_QUEUE_CLEANUP_INTERVAL	Task cleanup interval	30s
Authentication	A2A_AUTH_ENABLE	Enable OIDC authentication	false

Development

# Generate code from ADL
task generate

# Run tests
task test

# Build the application
task build

# Run linter
task lint

# Format code
task fmt

Adding Dependencies

The generator owns the baseline toolchain pins (SDK, server framework, logging, CLI, sandbox utilities). To extend the project without forking the templates, declare extras in agent.yaml - every empty list below is rendered by adl init --defaults precisely so it's discoverable:

Where	Purpose	Example entry	Rendered into
spec.language.go.vendor.deps	Runtime Go modules	github.com/stretchr/testify@v1.10.0	go.mod require block
spec.language.go.vendor.devdeps	Executable dev tools (Go 1.24 tool directive)	golang.org/x/tools/cmd/stringer@v0.20.0	go.mod tool directive
spec.development.deps	Cross-cutting sandbox tools (not tied to one language)	kubectl@1.31.0, terraform@1.9.5, deno@2.1.4	Flox manifest.toml / devcontainer feature

Entries use the <package>@<version> form. Built-in pins always win on conflict; the generator prints a warning and skips the user entry when shadowing is attempted. After editing agent.yaml, re-run task generate to refresh the manifests.

Debugging

Use the A2A Debugger to test and debug your A2A agent during development. It provides a web interface for sending requests to your agent and inspecting responses, making it easier to troubleshoot issues and validate your implementation.

docker run --rm -it --network host ghcr.io/inference-gateway/a2a-debugger:latest --server-url http://localhost:8080 tasks submit "What are your skills?"

docker run --rm -it --network host ghcr.io/inference-gateway/a2a-debugger:latest --server-url http://localhost:8080 tasks list

docker run --rm -it --network host ghcr.io/inference-gateway/a2a-debugger:latest --server-url http://localhost:8080 tasks get <task ID>

Deployment

Docker

The Docker image can be built with custom version information using build arguments:

# Build with default values from ADL
docker build -t documentation-agent .

# Build with custom version information
docker build \
  --build-arg VERSION=1.2.3 \
  --build-arg AGENT_NAME="My Custom Agent" \
  --build-arg AGENT_DESCRIPTION="Custom agent description" \
  -t documentation-agent:1.2.3 .

Available Build Arguments:

• VERSION - Agent version (default: 0.2.32)
• AGENT_NAME - Agent name (default: documentation-agent)
• AGENT_DESCRIPTION - Agent description (default: A2A agent server that provides Context7-style documentation capabilities for your agents)

These values are embedded into the binary at build time using linker flags, making them accessible at runtime without requiring environment variables.

License

Apache 2.0 License - see LICENSE file for details