Vercel AI SDK Documentation MCP Agent

A Model Context Protocol (MCP) server that provides AI-powered search and querying capabilities for the Vercel AI SDK documentation. This project enables developers to ask questions about the Vercel AI SDK and receive accurate, contextualized responses based on the official documentation.

MCP Compatible

TypeScript

Node.js

Features

Direct Documentation Search: Query the Vercel AI SDK documentation index directly using similarity search
AI-Powered Agent: Ask natural language questions about the Vercel AI SDK and receive comprehensive answers
Session Management: Maintain conversation context across multiple queries
Automated Indexing: Includes tools to fetch, process, and index the latest Vercel AI SDK documentation

Architecture

This system consists of several key components:

MCP Server: Exposes tools via the Model Context Protocol for integration with AI assistants
DocumentFetcher: Crawls and processes the Vercel AI SDK documentation
VectorStoreManager: Creates and manages the FAISS vector index for semantic search
AgentService: Provides AI-powered answers to questions using the Google Gemini model
DirectQueryService: Offers direct semantic search of the documentation

Setup Instructions

Prerequisites

Node.js 18+
npm
A Google API key for Gemini model access

Environment Variables

Create a .env file in the project root with the following variables:

GOOGLE_GENERATIVE_AI_API_KEY=your-google-api-key-here

You'll need to obtain a Google Gemini API key from the Google AI Studio.

Installation

Clone the repository

git clone https://github.com/IvanAmador/vercel-ai-docs-mcp.git
cd vercel-ai-docs-mcp-agent

Install dependencies
```
npm install
```
Build the project
```
npm run build
```
Build the documentation index
```
npm run build:index
```
Start the MCP server
```
npm run start
```

Integration with Claude Desktop

Claude Desktop is a powerful AI assistant that supports MCP servers. To connect the Vercel AI SDK Documentation MCP agent with Claude Desktop:

First, install Claude Desktop if you don't have it already.
Open Claude Desktop settings (via the application menu, not within the chat interface).
Navigate to the "Developer" tab and click "Edit Config".
Add the Vercel AI Docs MCP server to your configuration:

{
  "mcpServers": {
    "vercel-ai-docs": {
      "command": "node",  
      "args": ["ABSOLUTE_PATH_TO_PROJECT/dist/main.js"],
      "env": {
        "GOOGLE_GENERATIVE_AI_API_KEY": "your-google-api-key-here"
      }
    }
  }
}

Make sure to replace:

ABSOLUTE_PATH_TO_PROJECT with the actual path to your project folder
your-google-api-key-here with your Google Gemini API key

Save the config file and restart Claude Desktop.
To verify the server is connected, look for the hammer 🔨 icon in the Claude chat interface.

For more detailed information about setting up MCP servers with Claude Desktop, visit the MCP Quickstart Guide.

Integration with Other MCP Clients

This MCP server is compatible with any client that implements the Model Context Protocol. Here are a few examples:

Cursor

Cursor is an AI-powered code editor that supports MCP servers. To integrate with Cursor:

Add a .cursor/mcp.json file to your project directory (for project-specific configuration) or a ~/.cursor/mcp.json file in your home directory (for global configuration).
Add the following to your configuration file:

{
  "mcpServers": {
    "vercel-ai-docs": {
      "command": "node",  
      "args": ["ABSOLUTE_PATH_TO_PROJECT/dist/main.js"],
      "env": {
        "GOOGLE_GENERATIVE_AI_API_KEY": "your-google-api-key-here"
      }
    }
  }
}

For more information about using MCP with Cursor, refer to the Cursor MCP documentation.

Usage

The MCP server exposes three primary tools:

1. agent-query

Query the Vercel AI SDK documentation using an AI agent that can search and synthesize information.

{
  "name": "agent-query",
  "arguments": {
    "query": "How do I use the streamText function?",
    "sessionId": "unique-session-id"
  }
}

2. direct-query

Perform a direct similarity search against the Vercel AI SDK documentation index.

{
  "name": "direct-query",
  "arguments": {
    "query": "streamText usage",
    "limit": 5
  }
}

3. clear-memory

Clears the conversation memory for a specific session or all sessions.

{
  "name": "clear-memory",
  "arguments": {
    "sessionId": "unique-session-id"
  }
}

To clear all sessions, omit the sessionId parameter.

Development

Project Structure

├── config/              # Configuration settings
├── core/                # Core functionality
│   ├── indexing/        # Document indexing and vector store
│   └── query/           # Query services (agent and direct)
├── files/               # Storage directories
│   ├── docs/            # Processed documentation
│   ├── faiss_index/     # Vector index files
│   └── sessions/        # Session data
├── mcp/                 # MCP server and tools
│   ├── server.ts        # MCP server implementation
│   └── tools/           # MCP tool definitions
├── scripts/             # Build and utility scripts
└── utils/               # Helper utilities

Build Scripts

npm run build: Compile TypeScript files
npm run build:index: Build the documentation index
npm run dev:index: Build and index in development mode
npm run dev: Build and start in development mode

Troubleshooting

Common Issues

Index not found or failed to load

Run npm run build:index to create the index before starting the server.
API rate limits

When exceeding Google API rate limits, the agent service may return errors. Implement appropriate backoff strategies.
Model connection issues

Ensure your Google API key is valid and has access to the specified Gemini model.
Claude Desktop not showing MCP server
- Check your configuration file for syntax errors.
- Make sure the path to the server is correct and absolute.
- Check Claude Desktop logs for errors.
- Restart Claude Desktop after making configuration changes.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT