MCP Server

What is MCP Server?

MCP Server is a simple Node.js proxy server that exposes two API endpoints (GET /v1/models and POST /v1/chat/completions) for interacting with locally running LM Studio models. It bridges LM Studio and provides responses in an OpenAI-compatible format, intended for developers who need a lightweight middleware for local LLM access.

How to use MCP Server?

1. Clone the repository and run npm install to install dependencies.
2. Ensure LM Studio is running locally at http://localhost:11434.
3. Start the server with node index.js — it will listen on http://localhost:3000.
4. Use the two endpoints (GET /v1/models and POST /v1/chat/completions) to fetch models or send chat completions.

Key features of MCP Server

• Fetch available models from LM Studio via a GET endpoint
• Perform chat completions with selected models via a POST endpoint
• Built-in CORS support for cross-origin API calls
• Easy to extend and integrate into existing projects
• Returns responses in an OpenAI-compatible JSON format

Use cases of MCP Server

• Developing chat applications that run entirely on a local LLM
• Testing and comparing models served by LM Studio without a full API gateway
• Building lightweight frontend demos that require a simple proxy with CORS
• Creating automation scripts that need programmatic access to local LLM chats

FAQ from MCP Server

What are the runtime dependencies?

MCP Server requires Node.js, Express.js, Axios, Body-parser, and the CORS package. All listed in the README under “Environment Details”.

Does MCP Server require LM Studio to be running?

Yes. The server proxies requests to LM Studio at http://localhost:11434. LM Studio must be running on the same machine for endpoints to work.

Where does conversation data live?

All data stays local. The server runs on your machine and forwards requests to LM Studio, which also runs locally. No external service is involved.

What transport and authentication does MCP Server use?

The server uses standard HTTP (no TLS by default) and does not include authentication. It listens on http://localhost:3000 and expects clients to connect on the same host.

What are the known limitations of MCP Server?

The README does not list any specific limitations beyond the requirement that LM Studio must be running locally. The server provides only the two described endpoints and does not handle streaming, authentication, or advanced routing.