You run an MCP server that lets AI agents request help from other configured large language models through a unified API. This server provides multi-model access, a dynamic tool list, a simple memory of recent conversations per model, and easy environment-based configuration for your underlying API proxy. It’s designed to be integrated with your AI chat interface or MCP management tools so your代理s can query Grok, Gemini, Claude, GPT‑4o, and more through familiar OpenAI‑compatible requests.

How to use

After you start the MCP server in your client application, your AI agents will see a set of tools named like ask_grok3, ask_gemini2_5pro, ask_claude3_7sonnet, and ask_gpt4o. Each tool corresponds to a specific AI assistant. You provide a single prompt string to the tool, and the MCP server handles routing the request to the selected model, including built‑in memory of the last five interactions for context. The server uses your configured API proxy URL and API key to reach the backend models. Environment variables you set control where requests go and how they are authenticated.

How to install

Prerequisites: You need Node.js version 18.0.0 or higher.

node --version  # should output v18.0.0 or higher
```} ,{

Install and build from source using the following steps.

npm install

npm run build

Note: installation often runs the build automatically after npm install in typical setups.

## Additional notes and configuration

Configure the client application that launches this MCP server to provide the API URL and API key. The MCP server itself does not read a .env file; you supply these values to the client that starts the server.

Environment variables you should configure in your client setup:

- `API_URL` — the base URL of your OpenAI‑compatible API proxy (for example: `http://localhost:3000`). The server will append `/v1/chat/completions` to this base URL.
- `API_KEY` — your API key for accessing the proxy (for example: `sk-xxxxxxxx`).

Client configuration example (for a typical MCP client interface or plugin): use a stdio MCP channel with Node as the host process.

## Developer tips and testing

You can test the MCP server setup with the MCP Inspector tool if you have it installed. Ensure you set `API_URL` and `API_KEY` in the environment where you run the inspector.

npx @modelcontextprotocol/inspector build/index.js

# or if installed globally

mcp-inspector build/index.js

## Available tools

### ask\_grok3

Sends a prompt to the Grok3 model and returns its response.

### ask\_gemini2\_5pro

Queries the Gemini 2.5 Pro model for responses with context-aware reasoning.

### ask\_claude3\_7sonnet

Interacts with Claude 3.7 Sonnet variant to generate detailed answers.

### ask\_gpt4o

Requests responses from GPT-4o or compatible OpenAI‑style models.