OpenAI Agents SDK MCP Tool

A Model Context Protocol (MCP) server that provides documentation for the OpenAI Agents SDK by extracting and indexing content from the official documentation website: https://openai.github.io/openai-agents-python/

This package can be installed as a Python library and used with any MCP-compatible LLM client (Claude Desktop, VS Code, Cursor, etc.).

Installation

From PyPI (Recommended)

pip install openai-agents-sdk-mcp

From Source

git clone https://github.com/gavinz0228/openai-agents-sdk-mcp.git
cd openai-agents-sdk-mcp
pip install -e .

See INSTALLATION.md for detailed installation instructions.

Overview

This project provides both a standalone CLI tool and an MCP server that allows LLMs to access and query the OpenAI Agents SDK documentation intelligently.

Quick Start (MCP Server)

1. Install the package:

   pip install openai-agents-sdk-mcp
   

2. Set up API key:

   export OPENAI_API_KEY="sk-your-api-key-here"
   

3. Configure your MCP client (e.g., Claude Desktop):

   {
     "mcpServers": {
       "openai-agents-sdk-docs": {
         "command": "openai-agents-sdk-mcp",
         "env": {
           "OPENAI_API_KEY": "sk-your-api-key-here"
         }
       }
     }
   }
   

4. Start using it - Ask your LLM:

   • "List all OpenAI Agents SDK documentation topics"
   • "Get documentation for handoffs"
   • "How do I use streaming in OpenAI Agents?"

See INSTALLATION.md and MCP_CONFIGURATION.md for detailed setup instructions.

Features

1. MCP Server (Primary Interface)

Exposes two tools for LLMs:

• list_documentation_topics: Get a complete list of all available documentation topics with their URLs
• get_documentation: Search for and retrieve documentation using natural language queries

2. Automatic Documentation Indexing

• Fetches and parses the OpenAI Agents SDK documentation website
• Extracts all navigation links and topics
• Creates a structured JSON map of topics to URLs
• Saves to docs_index.json for quick access

2. Smart Index Management

The tool automatically manages the documentation index with intelligent caching:

• Missing Index Detection: Automatically fetches fresh index if docs_index.json doesn't exist
• Staleness Check: Refreshes index if older than 1 day (configurable)
• Link Validation: Verifies all documentation links are working
• Broken Link Recovery: Automatically re-fetches index if any links are broken

3. AI-Powered Feature Search

Uses OpenAI's GPT-4o-mini to intelligently match user queries to documentation:

• Accepts natural language queries (e.g., "how do I trace my agent")
• Finds the closest matching documentation topic
• Fetches and displays relevant documentation content
• Works with fuzzy matching and conversational queries

Installation

1. Clone the repository:

git clone https://github.com/gavinz0228/openai-agents-sdk-mcp.git
cd openai-agents-sdk-mcp

2. Install the package:

pip install -e .

3. Configure API key: Create a .env file in your working directory:

OPENAI_API_KEY=sk-your-api-key-here

Or set as environment variable:

export OPENAI_API_KEY="sk-your-api-key-here"

See INSTALLATION.md for more installation options.

Usage

MCP Server (Recommended)

The MCP server allows LLMs to access the documentation through standardized tool calls.

Start the Server

openai-agents-sdk-mcp

Or if running from source:

python -m openai_agents_sdk_mcp.server

Configure MCP Client

Add to your MCP client configuration (e.g., Claude Desktop's config):

{
  "mcpServers": {
    "openai-agents-sdk-docs": {
      "command": "openai-agents-sdk-mcp"
    }
  }
}

Or use the absolute path if installed in a virtual environment:

{
  "mcpServers": {
    "openai-agents-sdk-docs": {
      "command": "/path/to/.venv/bin/openai-agents-sdk-mcp"
    }
  }
}

Available MCP Tools

list_documentation_topics

• Lists all available documentation topics
• Optional parameter: force_refresh (boolean) - Force refresh the index

Example:

{
  "name": "list_documentation_topics",
  "arguments": {
    "force_refresh": false
  }
}

get_documentation

• Search and retrieve documentation for a specific feature
• Parameters:
  • query (string, required) - Feature name or natural language question
  • include_content (boolean, optional) - Whether to include full content (default: true)

Example:

{
  "name": "get_documentation",
  "arguments": {
    "query": "handoffs",
    "include_content": true
  }
}

Test the Server

python test_mcp.py

Command Line Interface

Use the CLI tool for quick documentation queries:

# List all documentation topics
openai-agents-docs

# Search for specific documentation
openai-agents-docs "handoffs"
openai-agents-docs "streaming"
openai-agents-docs "how to use guardrails"

As a Python Library

from openai_agents_sdk_mcp import (
    load_or_refresh_index,
    get_documentation_for_feature
)

# Load documentation index
doc_map = load_or_refresh_index()
print(f"Found {len(doc_map)} topics")

# Find documentation for a feature
topic, url = get_documentation_for_feature("handoffs")
if topic:
    print(f"Topic: {topic}")
    print(f"URL: {url}")

Standalone CLI Tool (Legacy)

If running from source without installation:

Generate/Refresh Documentation Index

python openai_agents_sdk_mcp.py

This will:

• Fetch the latest documentation structure
• Extract all topics and links
• Save to docs_index.json
• Display all available topics

Search for Documentation

python openai_agents_sdk_mcp.py "feature name or query"

Examples:

# Simple feature name
python openai_agents_sdk_mcp.py "handoffs"

# Natural language query
python openai_agents_sdk_mcp.py "how do I stream responses"

# Topic search
python openai_agents_sdk_mcp.py "tracing and debugging"

# Multiple words
python openai_agents_sdk_mcp.py "realtime voice"

The tool will:

1. Load or refresh the documentation index (if stale)
2. Use AI to find the best matching topic
3. Display the matched topic and URL
4. Fetch and show a preview of the documentation content

How It Works

Index Management

# The index is automatically managed:
# 1. Checks if docs_index.json exists
if not exists:
    fetch_fresh_index()

# 2. Checks if index is older than 1 day
if age > 1_day:
    fetch_fresh_index()

# 3. Validates all links are working
if broken_links_found:
    fetch_fresh_index()

AI-Powered Matching

The tool uses OpenAI's GPT-4o-mini to match user queries to documentation topics:

1. Loads all available topics from the index
2. Sends user query + topic list to the LLM
3. LLM identifies the most relevant topic
4. Returns the matching topic and URL

This provides intelligent matching even for:

• Typos and misspellings
• Natural language questions
• Partial or fuzzy matches
• Related concepts

Configuration

Constants (in openai_agents_sdk_mcp.py)

DOCS_INDEX_FILE = "docs_index.json"  # Index file name
INDEX_MAX_AGE_DAYS = 1               # Maximum age before refresh

Environment Variables

• OPENAI_API_KEY - Required for AI-powered search functionality

Files

• server.py - MCP server implementation
• openai_agents_sdk_mcp.py - Core functionality and CLI tool
• test_mcp.py - Test script for the MCP server
• mcp_config.json - Example MCP client configuration
• docs_index.json - Cached documentation index (auto-generated)
• requirements.txt - Python dependencies
• .env - Environment variables (create this)
• .gitignore - Git ignore rules (protects API key)

Dependencies

• requests - HTTP requests for fetching web pages
• beautifulsoup4 - HTML parsing
• lxml - XML/HTML parser
• openai - OpenAI API client for AI-powered search
• python-dotenv - Environment variable management
• mcp - Model Context Protocol SDK

Example Output

Index Generation

Fetching OpenAI Agents SDK documentation index...
Fetching fresh documentation index...
Index refreshed with 80 topics and saved to 'docs_index.json'.

Found 80 documentation topics/features:
...

Feature Search

Searching for documentation on: handoffs

Loaded existing index with 80 topics.
Verifying documentation links...
  ✓ All links are valid
✓ Found matching topic: Handoffs
  URL: https://openai.github.io/openai-agents-python/handoffs/

Fetching documentation content...
================================================================================
Handoffs - OpenAI Agents SDK
...

License

This project is designed to work with the OpenAI Agents SDK documentation. Please refer to OpenAI's terms of service for API usage.