mcp s

This MCP server provides an environment for Model Context Protocol tool calls within the LibreChat-based server experiment. It is designed to orchestrate MCP tools, run them in a sandboxed container, and expose a consistent API for the agent system to discover and execute tools. Once the container is running, you can interact with the MCP server to enumerate available tools, request tool executions, and receive results that can be used by higher-level agents or workflows within the LibreChat UI. The server supports tool invocation life-cycles, including input validation, sandboxed execution, and result streaming to the agent as appropriate.

To use the MCP server, connect your MCP-enabled agent to the server's tool registry endpoint exposed by the running container. Your agent can query the list of available tools, including metadata like tool name, description, required inputs, and output formats. When a tool is selected, the agent submits a call with the required inputs; the server executes the tool in a sandbox and returns the results and any side-effects (files, logs, etc.). This setup enables robust tool-calling workflows, with secure execution and clear separation between the agent logic and tool implementations.

Prerequisites:

• Docker installed and running on your host
• Basic familiarity with running containerized services

Step-by-step:

1. Ensure Docker is installed:

   • macOS: https://docs.docker.com/desktop/mac/
   • Windows: https://docs.docker.com/desktop/windows/
   • Linux: follow your distro’s instructions (e.g., apt/yum packages)

2. Pull and run the MCP server container (based on the repository configuration):

   • Run: docker pull alan-meigs/mcp-servers:latest
   • Start: docker run -d --name alan-meigs-mcp-servers -p 3000:3000 -e PORT=3000 -e MCP_LOG_LEVEL=info -e MCP_USE_SANDBOX=true alan-meigs/mcp-servers:latest

3. Verify the server is up:

   • Check logs: docker logs -f alan-meigs-mcp-servers
   • The server should listen on port 3000 inside the container and exposed to host at 3000

4. Integrate with LibreChat/MCP setup:

   • Ensure your MCP client configuration points to http://localhost:3000 (or the appropriate host if deploying remotely)
   • Use the MCP tooling APIs exposed by the server to enumerate and invoke tools

5. Optional: Use docker-compose (if provided by the repo) to manage multi-container setups. Create a docker-compose.yml with the service name alan-meigs-mcp-servers and include environment variables as needed, then run docker-compose up -d.

Environment variables can alter behavior of the MCP server. Common options include PORT (server port), MCP_LOG_LEVEL (e.g., debug, info, warn, error), and MCP_USE_SANDBOX (enable/disable sandboxed execution). If you modify port mappings, ensure your MCP clients are updated accordingly. If you encounter tool invocation failures, review container logs for sandbox permission messages, resource limits, or missing tool implementations. For production deployments, consider enabling TLS termination at your reverse proxy and restricting access to the MCP endpoints.

Troubleshooting tips:

• If the container fails to start, check for port conflicts on the host (try changing PORT).
• Ensure Docker has enough CPU/RAM resources allocated for sandboxed tool execution.
• Validate that the MCP tooling registry inside the container is populated with the expected tools before invoking them from an agent.

Configuration tips:

• Use MCP_USE_SANDBOX=true to enforce isolated execution environments for tool calls
• Set MCP_LOG_LEVEL=debug temporarily when diagnosing issues