🔍 GPT Researcher MCP Server
Why GPT Researcher MCP?
While LLM apps can access web search tools with MCP, GPT Researcher MCP delivers deep research results. Standard search tools return raw results requiring manual filtering, often containing irrelevant sources and wasting context window space.
GPT Researcher autonomously explores and validates numerous sources, focusing only on relevant, trusted and up-to-date information. Though slightly slower than standard search (~30 seconds wait), it delivers:
- ✨ Higher quality information
- 📊 Optimized context usage
- 🔎 Comprehensive results
- 🧠 Better reasoning for LLMs
💻 Claude Desktop Demo
https://github.com/user-attachments/assets/ef97eea5-a409-42b9-8f6d-b82ab16c52a8
Prerequisites
Before running the MCP server, make sure you have:
- Python 3.11 or higher installed
- Important: GPT Researcher >=0.12.16 requires Python 3.11+
- API keys for the services you plan to use:
You can also connect any other web search engines or MCP using GPTR supported retrievers. Check out the docs here
🚀 Running the MCP Server
You can run the MCP server in several ways:
Method 1: Directly using Python
python server.py
Method 2: Using the MCP CLI (if installed)
mcp run server.py
Method 3: Using Docker (recommended for production)
For n8n Integration
If you need to connect to an existing n8n network:
# First, start the container
docker-compose up -d
# Then connect to your n8n network
docker network connect n8n-mcp-net gptr-mcp
# Or create a shared network first
docker network create n8n-mcp-net
docker network connect n8n-mcp-net gptr-mcp
Note: The Docker image uses Python 3.11 to meet the requirements of gpt-researcher >=0.12.16. If you encounter errors during the build, ensure you're using the latest Dockerfile from this repository.
Once the server is running, you'll see output indicating that the server is ready to accept connections. You can verify it's working by:
- SSE Endpoint: Access the Server-Sent Events endpoint at http://localhost:8000/sse to get a session ID
- MCP Communication: Use the session ID to send MCP messages to http://localhost:8000/messages/?session_id=YOUR_SESSION_ID
- Testing: Run the test script with
python test_mcp_server.py
Important for Docker/n8n Integration:
- The server binds to
0.0.0.0:8000 to work with Docker containers
- Uses SSE transport for web-based MCP communication
- Session management requires getting a session ID from
/sse endpoint first
- Each client connection needs a unique session ID for proper communication
🚦 Transport Modes & Best Practices
The GPT Researcher MCP server supports multiple transport protocols and automatically chooses the best one for your environment:
Transport Types
| Transport | Use Case | When to Use |
|---|
| STDIO | Claude Desktop, Local MCP clients | Default for local development |
| SSE | Docker, Web clients, n8n integration | Auto-enabled in Docker |
| Streamable HTTP | Modern web deployments | Advanced web deployments |
Automatic Detection
The server automatically detects your environment:
# Local development (default)
python server.py
# ➜ Uses STDIO transport (Claude Desktop compatible)
# Docker environment
docker run gptr-mcp
# ➜ Auto-detects Docker, uses SSE transport
# Manual override
export MCP_TRANSPORT=sse
python server.py
# ➜ Forces SSE transport
Environment Variables
| Variable | Description | Default | Example |
|---|
MCP_TRANSPORT | Force specific transport | stdio | sse, streamable-http |
DOCKER_CONTAINER | Force Docker mode | Auto-detected | true |
Configuration Examples
For Claude Desktop (Local)
// ~/Library/Application Support/Claude/claude_desktop_config.json
{
"mcpServers": {
"gpt-researcher": {
"command": "python",
"args": ["/absolute/path/to/server.py"],
"env": {
"..."
}
}
}
}
For Docker/Web Deployment
# Set transport explicitly for web deployment
export MCP_TRANSPORT=sse
python server.py
# Or use Docker (auto-detects)
docker-compose up -d
For n8n MCP Integration
# Use the container name as hostname
docker run --name gptr-mcp -p 8000:8000 gptr-mcp
# In n8n, connect to: http://gptr-mcp:8000/sse
Transport Endpoints
When using SSE or HTTP transports:
- Health Check:
GET /health
- SSE Endpoint:
GET /sse (get session ID)
- MCP Messages:
POST /messages/?session_id=YOUR_SESSION_ID
Best Practices
- Local Development: Use default STDIO for Claude Desktop
- Production: Use Docker with automatic SSE detection
- Testing: Use health endpoints to verify connectivity
- n8n Integration: Always use container networking with Docker
- Web Deployment: Consider Streamable HTTP for modern clients
Integrating with Claude
You can integrate your MCP server with Claude using:
Claude Desktop Integration - For using with Claude desktop application on Mac
For detailed instructions, follow the link above.
💻 Claude Desktop Integration
To integrate your locally running MCP server with Claude for Mac, you'll need to:
- Make sure the MCP server is installed and running
- Configure Claude Desktop:
- Locate or create the configuration file at
~/Library/Application Support/Claude/claude_desktop_config.json
- Add your local GPT Researcher MCP server to the configuration with environment variables
- Restart Claude to apply the configuration
⚠️ Important: Environment Variables Required
Claude Desktop launches your MCP server as a separate subprocess, so you must explicitly pass your API keys in the configuration. The server cannot access your shell's environment variables or .env file automatically.
Configuration Example
{
"mcpServers": {
"gptr-mcp": {
"command": "python",
"args": ["/absolute/path/to/your/server.py"],
"env": {
"OPENAI_API_KEY": "your-actual-openai-key-here",
"TAVILY_API_KEY": "your-actual-tavily-key-here"
}
}
}
}
Security Note
🔒 Your Claude Desktop config contains sensitive API keys. Protect it:
chmod 600 ~/Library/Application\ Support/Claude/claude_desktop_config.json
Never commit this file to version control.
Alternative: Environment Variable Script
For better security, create a wrapper script:
run_gptr_mcp.sh:
#!/bin/bash
source /path/to/your/.env
python /absolute/path/to/server.py
Then use it in Claude Desktop:
{
"mcpServers": {
"gptr-mcp": {
"command": "/absolute/path/to/run_gptr_mcp.sh"
}
}
}
For complete step-by-step instructions, see the Claude Desktop Integration guide.
🔧 Troubleshooting
If you encounter issues while running the MCP server:
General Issues
- API Keys: Make sure your API keys are correctly set in the
.env file
- Python Version: Check that you're using Python 3.11 or higher (required by gpt-researcher >=0.14.0)
- Dependencies: Ensure all dependencies are installed correctly:
pip install -r requirements.txt
- Server Logs: Check the server logs for error messages
Docker Issues
-
Container not accessible:
- Verify the container is running:
docker ps | grep gptr-mcp
- Check container logs:
docker logs gptr-mcp
- Confirm the server is binding to 0.0.0.0:8000 (logs should show this)
-
n8n Integration Issues:
- Ensure both containers are on the same Docker network
- Use the container name
gptr-mcp as the hostname in n8n
- Set the MCP server URL to:
http://gptr-mcp:8000/sse
-
Session ID Issues:
- The server uses SSE transport which requires session management
- First, get a session ID by connecting to
/sse endpoint
- Use the session ID in subsequent MCP requests:
/messages/?session_id=YOUR_ID
- Each client needs its own session ID
n8n MCP Integration Steps
-
Get Session ID:
curl http://gptr-mcp:8000/sse
# Look for: data: /messages/?session_id=XXXXX
-
Initialize MCP:
curl -X POST http://gptr-mcp:8000/messages/?session_id=YOUR_SESSION_ID \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "id": 1, "method": "initialize", "params": {"protocolVersion": "2024-11-05", "capabilities": {"roots": {"listChanged": true}}, "clientInfo": {"name": "n8n-client", "version": "1.0.0"}}}'
-
Call Tools:
curl -X POST http://gptr-mcp:8000/messages/?session_id=YOUR_SESSION_ID \
-H "Content-Type: application/json" \
-d '{"jsonrpc": "2.0", "id": 2, "method": "tools/call", "params": {"name": "quick_search", "arguments": {"query": "test"}}}'
Testing the Server
Run the included test script to verify functionality:
python test_mcp_server.py
This will test:
- SSE connection and session ID retrieval
- MCP initialization
- Tool discovery and execution
Claude Desktop Issues
If your MCP server isn't working with Claude Desktop:
-
Server not appearing in Claude:
- Check your
claude_desktop_config.json syntax is valid JSON
- Ensure you're using absolute paths (not relative)
- Verify the path to
server.py is correct
- Restart Claude Desktop completely
-
"OPENAI_API_KEY not found" error:
- Make sure you added API keys to the
env section in your config
- Don't forget both
OPENAI_API_KEY and TAVILY_API_KEY
- API keys should be the actual keys, not placeholders
-
Tools not showing up:
- Look for the 🔧 tools icon in Claude Desktop
- Check that Claude Desktop config file is in the right location:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json
- Windows:
%APPDATA%\Claude\claude_desktop_config.json
-
Python/Permission issues:
- Make sure Python is accessible from the command line:
python --version
- Try using full Python path:
"command": "/usr/bin/python3" or "command": "python3"
- Check file permissions on your server.py file
-
Still not working?
- Test the server manually:
python server.py (should show STDIO transport message)
👣 Next Steps
📄 License
This project is licensed under the MIT License - see the LICENSE file for details.
📞 Support / Contact