Mistral MCP server example

This is a TypeScript-based MCP server that provides two tools for chatting with Mistral. It is a basic example of how to create a server that can be used with the Model Context Protocol (MCP).

Blog post

This repository is part of a blog post by Speakeasy: Building an MCP server for Mistral.

Requirements

Node.js (tested on v20.17.10)
An MCP client that supports tools. We recommend the Claude desktop application or the Cline VSCode extension.

Mistral API key

To interact with the Mistral AI platform, you'll need an API key. You can get one by signing up at mistral.ai.

Set your API key as an environment variable. Create a copy of .env.example and rename it .env.

cp .env.example .env
open .env

Now update the .env file with your Mistral API key:

MISTRAL_API_KEY="YOUR_MISTRAL_API_KEY"

Tools

This server provides two tools.

For chatting with Mistral using text input, the mistral_chat_text tool:

Takes a model and an array of text inputs
Returns a text response from Mistral.

For chatting with Mistral using text and image input, the mistral_chat_image tool:

Takes a model and an array of text and image inputs
Only accepts images hosted on a public URL
Returns a text response from Mistral

Development

Install dependencies:

npm install

Build the server:

npm run build

For development with auto-rebuild:

npm run watch

Installation

To use this server with the Claude desktop app, add the following server config to your Claude config file:

On MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "Mistral MCP Server": {
      "command": "node",
      "args": [
        // Update this path to the location of the built server
        "/Users/speakeasy/server-mistral/build/index.js"
      ],
      "env": {
        // Update this with your Mistral API key
        "MISTRAL_API_KEY": "YOUR_MISTRAL_API_KEY"
      }
    }
  }
}

Debugging

Since MCP servers communicate through standard input/output streams (stdio), debugging can be challenging. We recommend using the MCP Inspector, which is available as a package script:

npm run inspector

The MCP Inspector will provide a URL for accessing debugging tools in your browser.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Mistral MCP Server

Integrations