Documentation & context fetcher for Dart & Flutter packages
Agent-agnostic. Pub.dev integrated. Zero-cloning required.

Installation • Tools • Architecture • Client Setup • Testing

---

Your AI coding agent knows the language. dart_docs_mcp gives it the library knowledge it's missing.

A Dart-based MCP server that fetches metadata, READMEs, examples, and API surfaces directly from pub.dev and GitHub. Works with any agent that supports MCP — Gemini CLI, Antigravity, Claude Code, Cursor, or anything else.

Features

• 📦 Metadata & Docs: Fetches READMEs and example/ directories for instant library onboarding.
• 🔍 API Surface Exploration: Provides a concise "virtual header" of public classes and methods.
• ⚡ Source Grepping: Search internal lib/ files using regex without cloning the repository.
• 📈 Changelog Intelligence: Extracts targeted version ranges to assist with migrations and breaking changes.
• 🌳 Type Hierarchy: Reconstructs inheritance trees for complex class structures.

Installation

Via Homebrew (macOS & Linux)

brew tap alfredobs97/tap
brew install dart-docs-mcp

Running Locally (Development)

dart run bin/server.dart

Using Docker

# Production AOT build
docker build --target runtime -t dart-docs-mcp:latest .
docker run -i dart-docs-mcp:latest

Tools Exposed

Tool	Capability
get_package_docs	Fetch README + example/ for usage context.
get_api_surface	Get a "virtual header" of all public APIs from index.json.
search_package_code	Grep internal source code via keyword/regex (capped at 15 matches).
get_changelog	Targeted version diffs between specific releases.
get_type_hierarchy	Map class inheritance and implementations (e.g., sealed classes).
cross_reference	Map README features to specific implementation files.

Architecture

This server is built using the Model Context Protocol (MCP), an open standard that enables AI models to interact with local and remote tools.

1. Request: An MCP client sends a JSON-RPC request over stdio.
2. Orchestration: The server routes requests to specific tool implementations in lib/mcp/tools/.
3. Execution: Services fetch data from pub.dev API, Dartdoc assets, and GitHub fallbacks.
4. Response: Context is formatted and injected back into the LLM's conversation.

Adding to an MCP Client

Gemini CLI

gemini mcp add local stdio dart_docs_mcp --command dart-docs-mcp

Google Antigravity IDE

Add this to your MCP configuration JSON:

{
  "mcpServers": {
    "dart_docs_mcp": {
      "command": "dart-docs-mcp",
      "args": []
    }
  }
}

Claude Code

claude mcp add dart_docs_mcp -- dart-docs-mcp

Local Testing & Debugging

Use the mcp_dart_cli inspector to call tools manually:

# Install the CLI
dart pub global activate mcp_dart_cli

# Test a tool query
mcp_dart inspect --tool get_package_docs --json-args '{"package_name": "http"}'

Testing

dart test