MCP Server for LangGraph Agent
FastMCP-based Model Context Protocol server for ChatGPT Enterprise integration with LangGraph agent.
🎉 Quick Start
The fastest way to get started:
./start.sh
This will:
- Check if your LangGraph agent is running on port 2024
- Start the MCP server on port 8000
- Start the test UI on port 3005
- Open http://localhost:3005 in your browser
Overview
This MCP server provides a standardized interface to interact with a LangGraph agent deployed on port 2024, compliant with ChatGPT Enterprise integration requirements.
✨ Latest Updates
January 2025 - LangGraph CLI Integration
- ✅ Refactored to use LangGraph CLI API architecture
- ✅ Updated to
/runs,/runs/stream,/okendpoints - ✅ Changed message format to
{"type": "human"} - ✅ Added 3 new tools: health check, agent status, thread listing
- ✅ Fixed all async operations (no more blocking calls)
- ✅ Updated web UI to match new API structure
See REFACTORING_SUMMARY.md for detailed changes.
Features
- ✅ MCP Protocol 2025-06-18 compliant
- ✅ ChatGPT Enterprise compatible (SSE transport)
- ✅ OAuth 2.0 Authentication - Google OAuth and API key support
- ✅ FastMCP 2.13.0+ framework for production-ready deployment
- ✅ LangGraph CLI API integration
- ✅ 6 Tools: invoke_agent, stream_agent, check_system_health, check_agent_status, get_thread_state, list_threads
- ✅ 2 Resources: Agent health check and server info
- ✅ Prompts: Formatted agent queries
- ✅ Web Test UI: Interactive testing interface on port 3005
- ✅ Secure by Default: Optional authentication for production deployments
Installation
- Install dependencies:
pip install -r requirements.txt
- Configure environment (optional):
# Copy example configuration
cp .env.example .env
# Edit .env with your settings
# For development without auth:
OAUTH_ENABLED=false
# For production with auth:
OAUTH_ENABLED=true
GOOGLE_CLIENT_ID=your-client-id
GOOGLE_CLIENT_SECRET=your-client-secret
API_KEYS=your-api-key-1,your-api-key-2
- Generate credentials (if using OAuth):
python generate_credentials.py
See OAUTH_SETUP.md for detailed authentication setup.
Usage
Option 1: Quick Start Script (Recommended)
./start.sh
Option 2: Manual Start
Start MCP Server:
python src/agent_mcp/mcp_server.py
Start Test UI (optional):
cd web_ui && python server.py
Option 3: Using FastMCP CLI
python -m agent_mcp.mcp_server
Or using FastMCP CLI:
fastmcp run src/agent_mcp/mcp_server.py
For local development (STDIO):
fastmcp dev src/agent_mcp/mcp_server.py
Custom transport:
from agent_mcp.mcp_server import mcp
# HTTP transport
mcp.run(transport="http", host="0.0.0.0", port=8000, path="/mcp")
# SSE transport (for ChatGPT Enterprise)
mcp.run(transport="sse", host="0.0.0.0", port=8000)
Available Tools
1. invoke_agent
Execute a single invocation of the LangGraph agent.
{
"prompt": "What is the capital of France?",
"thread_id": "optional-thread-id"
}
2. stream_agent
Stream responses from the LangGraph agent.
{
"prompt": "Tell me a story",
"thread_id": "optional-thread-id"
}
3. get_agent_state
Retrieve the current state of a conversation thread.
{
"thread_id": "thread-id-to-query"
}
Authentication
The MCP server supports three authentication methods for production deployments:
1. OAuth 2.0 (Google or Okta)
Enable user-based authentication with your preferred identity provider:
Google OAuth:
# .env configuration
OAUTH_ENABLED=true
OAUTH_PROVIDER=google
GOOGLE_CLIENT_ID=your-client-id.apps.googleusercontent.com
GOOGLE_CLIENT_SECRET=your-client-secret
Okta OAuth:
# .env configuration
OAUTH_ENABLED=true
OAUTH_PROVIDER=okta
OKTA_DOMAIN=your-domain.okta.com
OKTA_CLIENT_ID=your-okta-client-id
OKTA_CLIENT_SECRET=your-okta-client-secret
OAuth Endpoints:
GET /auth/login- Initiate OAuth flowGET /auth/callback- OAuth callback handlerGET /auth/logout- LogoutGET /auth/status- Check authentication status
Quick Start Guides:
2. API Key Authentication
Use API keys for service-to-service authentication:
# Generate API keys
python generate_credentials.py
# Add to .env
API_KEYS=key1,key2,key3
Using API Keys:
# cURL
curl -H "X-API-Key: your-api-key" http://localhost:8000/sse
# Python
headers = {"X-API-Key": "your-api-key"}
Testing Authentication
# Test OAuth setup
python test_oauth.py
# Or test manually
curl http://localhost:8000/health # Public endpoint
curl -H "X-API-Key: your-key" http://localhost:8000/sse # Protected
For detailed setup instructions, see OAUTH_SETUP.md
Resources
agent://health- Agent health checkagent://info- Agent capabilities and metadata
Prompts
agent_query_prompt- Format queries for the agent
ChatGPT Enterprise Integration
This server is designed for ChatGPT Enterprise integration with:
- SSE Transport: Default transport for real-time streaming
- MCP Protocol 2025-06-18: Latest stable protocol version
- Proper Tool Schemas: Auto-generated from Python type hints
- Context Support: Logging and progress reporting
- Error Handling: Comprehensive error responses
📚 Enterprise Apps & Connectors Documentation
Having trouble integrating with ChatGPT Enterprise "Apps & Connectors"?
We've created comprehensive documentation to help you:
- Start Here: Executive Summary - Quick overview of the issue and solutions
- Full Research Report - Detailed technical analysis
- Troubleshooting Checklist - Step-by-step diagnostic guide
- Code Fixes Guide - Ready-to-implement improvements
- Enterprise Admin Guide - For workspace administrators
- Visual Architecture Guide - Diagrams and flow charts
- Complete Documentation Index - All resources in one place
TL;DR: If your OpenAPI server works in Custom GPTs but not in Enterprise "Apps & Connectors", it's likely due to Enterprise workspace domain restrictions. Contact your Enterprise admin to whitelist mcp.marananapp.com in the GPT Actions settings.
ChatGPT Configuration
Add to your ChatGPT Enterprise MCP configuration:
{
"mcpServers": {
"langgraph-agent": {
"url": "http://your-server:8000/sse",
"transport": "sse"
}
}
}
Testing
Web UI Test Tool
We provide a beautiful web-based UI to test your MCP server and LangGraph agent:
# Start the test UI server
cd web_ui
python server.py
Then open http://localhost:3005 in your browser to:
- Test MCP server connectivity
- Test LangGraph agent connectivity
- Invoke agent with custom prompts
- Stream responses in real-time
- View activity logs
See web_ui/README.md for details.
Unit Tests
Run tests:
pytest tests/test_mcp_server.py -v
Run all tests:
pytest
Development
Project Structure
agent-mcp-py/
├── src/
│ └── agent_mcp/
│ ├── __init__.py
│ └── mcp_server.py # FastMCP server implementation
├── tests/
│ ├── test_mcp_server.py # MCP server tests
│ └── test_*.py # Other tests
├── requirements.txt # Production dependencies
├── requirements-dev.txt # Development dependencies
└── README.md
Adding New Tools
from fastmcp import Context
@mcp.tool()
async def my_tool(param: str, ctx: Context = None) -> dict:
"""Tool description for ChatGPT."""
if ctx:
await ctx.info(f"Processing: {param}")
# Your logic here
return {"result": "success"}
Adding Resources
@mcp.resource("custom://resource")
async def my_resource() -> str:
"""Resource description."""
return "Resource content"
Architecture
┌─────────────────┐
│ ChatGPT │
│ Enterprise │
└────────┬────────┘
│ MCP/SSE
│
┌────────▼────────┐
│ FastMCP │
│ Server │
│ (Port 8000) │
└────────┬────────┘
│ HTTP
│
┌────────▼────────┐
│ LangGraph │
│ Agent │
│ (Port 2024) │
└─────────────────┘
Production Deployment
With Authentication
from fastmcp.server.auth.providers.google import GoogleProvider
auth = GoogleProvider(
client_id="your-client-id",
client_secret="your-client-secret",
base_url="https://your-domain.com"
)
mcp = FastMCP(
"LangGraph Agent Server",
auth=auth
)
Docker Deployment
FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY src/ ./src/
CMD ["python", "-m", "agent_mcp.mcp_server"]
License
MIT