Zerodha Kite Connect MCP

An MCP server that exposes Kite Connect's Python SDK operations as tools. REST methods run in worker threads so synchronous SDK requests do not block the MCP event loop. KiteTicker is controlled through dedicated WebSocket tools.

Setup

Requires Python 3.10 or newer.

cd D:\PRO\Zerodha_MCP
py -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install -r requirements.txt
Copy-Item .env.example .env

Set KITE_API_KEY and KITE_API_SECRET in .env. The default callback is http://127.0.0.1:5000/login; set KITE_REDIRECT_URL if the redirect URL registered in the Kite developer console is different. The URL must match exactly, including host, port, and path. Keep .env private; it is ignored by Git.

The MCP starts a Flask loopback callback listener on that URL using the flow in login.py. Call kite_login to get a Zerodha login link. After sign-in, Kite redirects to /login with a short-lived request_token; the MCP exchanges it for an access token, updates the running client, and persists the token to .env. The browser page reports success without displaying the token. If any authenticated SDK tool gets a Kite TokenException, it returns an authentication_required result with a fresh login link. Complete login, then retry the original tool. The direct generate_session and set_access_token tools also persist refreshed tokens.

Start using stdio:

python server.py

Example MCP client configuration:

{
  "mcpServers": {
    "zerodha-kite-connect": {
      "command": "D:\\PRO\\Zerodha_MCP\\.venv\\Scripts\\python.exe",
      "args": ["D:\\PRO\\Zerodha_MCP\\server.py"],
      "env": {
        "KITE_API_KEY": "your-api-key",
        "KITE_API_SECRET": "your-api-secret",
        "KITE_REDIRECT_URL": "http://127.0.0.1:5000/login"
      }
    }
  }
}

Related MCP server: Trade-MCP

Tools

Every REST SDK method has its own tool name. Its SDK keyword parameters are passed inside an arguments JSON object. Examples include place_order, orders, order_history, order_trades, positions, holdings, quote, ohlc, ltp, instruments, historical_data, margin calculations, GTT, mutual fund orders and SIPs, session login/token lifecycle, and auction tools. The exact accepted parameters and values are those of the installed kiteconnect SDK; Kite Connect's own validation and errors are returned.

The ticker_* tools manage the KiteTicker WebSocket: connect, subscribe, unsubscribe, set mode, check connection, read recent ticks, resubscribe, stop retries, stop, and close. Call ticker_connect before subscribing. Recent ticks are kept in a bounded in-memory buffer (up to 5,000) and retrieved with ticker_read_ticks.

SQLite audit batching

Every tool invocation is queued in memory, then written with SQLite executemany from a background worker. The worker flushes at the configured batch size, after the flush interval for partial batches, and on clean shutdown. Defaults are 25 calls or 2 seconds, whichever comes first.

The tool_calls table stores UTC time, tool name, JSON arguments and result, error text, and duration in milliseconds. Credential fields such as access, refresh, request, and API secret tokens are redacted. Large JSON values are truncated to keep the audit database bounded per call. Since batching is in-memory, an abrupt process termination can lose the pending batch.

Notes

• This server sends real requests to the authenticated Kite account. Order placement, modification, cancellation, GTT, and mutual fund tools can cause financial transactions.

• The registered redirect URL must be reachable on the same machine running this MCP process. Port 5000 must be free when using the default URL.

• Kite redirects with a request_token, not an access code. The callback exchanges it server-side; the API secret and access token are not shown in the browser response.

• Ticker and REST methods require an active Kite session. The daily session lifecycle and API permissions are controlled by Zerodha.