Context7 Chat MCP Server

An MCP (Model Context Protocol) server that provides AI-powered answers about library documentation using Context7's documentation database.

What it does

This MCP server exposes a single tool (query-docs) that allows LLMs to ask questions about any library and get formatted, up-to-date answers based on Context7's documentation database.

Unlike the main Context7 MCP server which retrieves raw documentation, this server returns AI-generated responses that directly answer your questions.

Requirements

• Node.js >= v18.0.0
• Context7 API Key (required) - Get yours at context7.com/dashboard

Usage

Adding to Cursor

Add to your ~/.cursor/mcp.json:

{
  "mcpServers": {
    "context7-chat": {
      "command": "npx",
      "args": ["-y", "@upstash/context7-chat-mcp", "--api-key", "YOUR_API_KEY"]
    }
  }
}

Adding to Claude Code

claude mcp add context7-chat -- npx -y @upstash/context7-chat-mcp --api-key YOUR_API_KEY

Adding to VS Code

Add to your VS Code MCP settings:

{
  "mcp": {
    "servers": {
      "context7-chat": {
        "type": "stdio",
        "command": "npx",
        "args": ["-y", "@upstash/context7-chat-mcp", "--api-key", "YOUR_API_KEY"]
      }
    }
  }
}

Testing with MCP Inspector

npx @modelcontextprotocol/inspector npx -y @upstash/context7-chat-mcp --api-key YOUR_API_KEY

Available Tools

query-docs

Ask a question about any library and get an AI-powered answer.

Parameters:

| Parameter | Type | Required | Description | | --------- | ------ | -------- | --------------------------------------------------------------- | | query | string | Yes | Your question about a library | | library | string | No | Library to focus the search on (recommended for better results) |

Examples:

query: "How do I create a checkout session?"
library: "stripe"

query: "How to set up server-side rendering?"
library: "nextjs"

query: "What are React hooks and how do I use useState?"

Configuration Options

| Option | Description | | ----------- | ------------------------------------------------------------------------------------------------ | | --api-key | Your Context7 API key (required) | | --model | OpenRouter model ID to use for generating responses (e.g., google/gemini-2.5-flash) (optional) |

Environment Variables

| Variable | Description | | ------------------ | ------------------------------------------------------- | | CONTEXT7_API_KEY | Your Context7 API key (alternative to --api-key flag) |

Using a Custom Model

You can specify which OpenRouter model to use for generating responses:

npx -y @upstash/context7-chat-mcp --api-key YOUR_API_KEY --model google/gemini-2.5-flash

Cursor configuration with custom model:

{
  "mcpServers": {
    "context7-chat": {
      "command": "npx",
      "args": [
        "-y",
        "@upstash/context7-chat-mcp",
        "--api-key",
        "YOUR_API_KEY",
        "--model",
        "google/gemini-2.5-flash"
      ]
    }
  }
}

Local Development

Clone the repository and navigate to the package:

git clone https://github.com/upstash/context7.git
cd context7/packages/mcp-chat

Install dependencies and build:

pnpm install
pnpm run build

Run the server:

# Using built version
node dist/index.js --api-key YOUR_API_KEY

# Using TypeScript source directly
npx tsx src/index.ts --api-key YOUR_API_KEY

Local MCP Configuration

For local development, you can configure your MCP client to use the TypeScript source directly:

Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "context7-chat": {
      "command": "npx",
      "args": [
        "tsx",
        "/path/to/context7/packages/mcp-chat/src/index.ts",
        "--api-key",
        "YOUR_API_KEY"
      ]
    }
  }
}

VS Code:

{
  "mcp": {
    "servers": {
      "context7-chat": {
        "type": "stdio",
        "command": "npx",
        "args": [
          "tsx",
          "/path/to/context7/packages/mcp-chat/src/index.ts",
          "--api-key",
          "YOUR_API_KEY"
        ]
      }
    }
  }
}

Development Commands

# Install dependencies
pnpm install

# Build
pnpm run build

# Watch mode (auto-rebuild on changes)
pnpm run dev

# Run built version
pnpm run start

# Lint
pnpm run lint

# Format
pnpm run format

License

MIT