kite

Kite MCP Server exposes a unified MCP interface that lets AI agents securely access the Kite Connect trading API. It aggregates market data, portfolio management, and order execution endpoints into a consistent, model-context-friendly protocol. The server can run in HTTP (default), SSE, or stdio modes, enabling flexible deployment for hosted or self-hosted setups. To use it, point your MCP client (or Claude Desktop via mcp-remote) at the server’s /mcp endpoint in HTTP mode, and request the available tools such as get_quotes, place_order, get_holdings, and more. The server also serves runtime documentation, making it easier for agents to discover and invoke the Kite Connect endpoints they need.

Key capabilities include real-time market data access (quotes, OHLC, historical data), portfolio and margin visibility, comprehensive order management (place, modify, cancel, and history), GTT (Good Till Triggered) order handling, and paginated responses for large datasets. In addition to core trading functionality, the MCP server includes setup helpers and example configurations for easy integration with Claude Desktop and other MCP clients via hosted or self-hosted endpoints.

To use the tools, invoke the corresponding MCP commands exposed by the server (e.g., get_quotes, get_holdings, place_order, get_orders, get_trades, get_gtts, etc.) through your AI agent or client. When running self-hosted, ensure your Kite Connect credentials are provided via environment variables (KITE_API_KEY and KITE_API_SECRET) and that APP_MODE is set to http, sse, or stdio as needed. If running in hosted HTTP mode, you can rely on the hosted endpoint and simply configure your client to connect to the provided /mcp endpoint.

Prerequisites:

• Go 1.21 or later installed on your system
• Git installed
• Access to Kite Connect API credentials (KITE_API_KEY and KITE_API_SECRET) for self-hosting

1. Install Go toolchain (if not already):

   • macOS: brew install go
   • Linux: sudo apt-get install golang-go
   • Windows: download and install from golang.org

2. Clone the repository: git clone https://github.com/zerodha/kite-mcp-server.git cd kite-mcp-server

3. Build the server binary (example for HTTP mode): go build -o kite-mcp-server

4. Create a .env or configure environment variables for self-hosting: APP_MODE=http APP_PORT=8080 APP_HOST=localhost KITE_API_KEY=your_api_key KITE_API_SECRET=your_api_secret

5. Run the server: ./kite-mcp-server

6. Verify the server is running by visiting http://localhost:8080/ in a browser (status page and self-hosted docs should be available).

Tips and notes:

• For Claude Desktop or other MCP clients, you can run the server in HTTP mode and point the client to the /mcp endpoint (or /sse for SSE mode).
• If you’re using hosted Kite data, you can rely on the hosted endpoint; for self-hosting, supply Kite credentials via KITE_API_KEY and KITE_API_SECRET.
• In Claude Desktop configurations, you’ll typically set command to a runner (e.g., npx mcp-remote) and provide the appropriate URL to the MCP endpoint along with APP_MODE and Kite credentials in env.
• Ensure proper networking: when running on localhost, port 8080 must be reachable by the client. If deploying behind a reverse proxy or load balancer, configure the proxy to forward /mcp accordingly.
• The server supports automatic pagination for large datasets (holdings, orders, trades); be mindful of rate limits and implement sensible query patterns in your agent.