# Jina AI MCP Server

A remote MCP (Model Context Protocol) server deployed on Cloudflare Workers that wraps Jina AI's APIs as MCP tools.

## Tools

| Tool | Jina API | Description |
|------|----------|-------------|
| `read_url` | Reader API (`r.jina.ai`) | Fetch any URL and return content as markdown/text/html |
| `search_web` | Search API (`s.jina.ai`) | Web search returning titles, URLs, and snippets |
| `embed_text` | Embeddings API (`api.jina.ai/v1/embeddings`) | Generate vector embeddings for text |
| `rerank` | Reranker API (`api.jina.ai/v1/rerank`) | Rerank documents by relevance to a query |
| `classify` | Classifier API (`api.jina.ai/v1/classify`) | Zero-shot text classification |

## Setup

### 1. Install dependencies

```bash
npm install
```

### 2. Set your Jina API key as a Worker secret

```bash
npx wrangler secret put JINA_API_KEY
```

Paste your Jina API key when prompted. It will be stored encrypted and injected as `env.JINA_API_KEY` at runtime.

### 3. Deploy

```bash
npm run deploy
```

Your MCP server will be live at:

```
https://jina-mcp-server.<your-subdomain>.workers.dev/mcp
```

### 4. Connect from an MCP client

Add the server URL to any MCP-compatible client (Claude Desktop, Cursor, Windsurf, AI Playground, MCP Inspector, etc.):

```json
{
  "mcpServers": {
    "jina": {
      "url": "https://jina-mcp-server.<your-subdomain>.workers.dev/mcp"
    }
  }
}
```

## Local development

```bash
echo 'JINA_API_KEY=your_key_here' > .dev.vars
npm run dev
```

The server will be available at `http://localhost:8787/mcp`.