mcp -manager

This MCP server manager is a Python-based orchestration layer that automatically wraps stdio-based MCP servers with Supergateway to expose them as SSE endpoints and provides a web inspector for testing. It supports auto-updating dependencies, background monitoring to restart crashed servers, and a web UI for debugging MCP endpoints via browser. The example configuration shows how to add a server (here, the GitHub MCP server) with its command, arguments, environment variables, port, and server type. To use, configure your MCP server entries, start the servers with the provided Make targets, and then connect to their SSE endpoints through the inspector or your client tooling. Tools available include the web inspector for testing MCP endpoints, automatic updates of npm packages and git repos before startup, port conflict handling, and automatic crash recovery with logging to /tmp for each server.

Prerequisites:

• Python 3.8+ and pip
• Node.js and npm (if you plan to run any Node-based servers or client tooling)
• Git (for pulling repositories)

Installation steps:

1. Clone the repository or download the package.
2. Install Python dependencies:

   pip install -r requirements.txt
   

3. Copy environment example and set API keys and tokens:

   cp .env.example .env
   # Edit .env to add your API keys as needed
   

4. Start all MCP servers with auto-restart:

   make run-servers
   

5. Start the web inspector for debugging:

   make run-inspector
   

6. Open the inspector URL shown in the console and use the UI to test MCP endpoints.

Tip: If you add additional MCP servers, update mcp_config.json with the new server definitions and restart the manager.

Environment variables are loaded using the ${VAR} syntax from the .env file. Common issues include port conflicts (resolved by make check-server-ports and make kill-server-conflicts), and authentication tokens expiring (update the token in the .env file or the specific server's env block). Logs for each server are written to /tmp/<server_name>.log, which helps diagnose startup failures or runtime crashes. The manager will attempt to restart crashed servers up to 3 times, and it can be configured to keep servers running without auto-restart by using KEEP_ALIVE=0 when starting. When adding a new MCP server definition, ensure that the port does not collide with existing services and that the command/args format matches the server type (stdio, legacy, etc.).