mcpField
K
Kite Connect MCP

Kite Connect MCP

Local
github.com/techfreakwormFinance0Updated 5d ago

About Kite Connect MCP

Kite Connect MCP is a self-hostable MCP server for connecting AI assistants to the Zerodha Kite Connect v3 API. It covers account, market and trading operations, with writes defaulting to dry runs and configurable safeguards for live actions.

Key features

  1. 01Account, portfolio and order tools
  2. 02Quotes, historical data and live ticker
  3. 03GTTs, alerts and mutual fund operations
  4. 04Dry-run-first writes with confirmation
  5. 05Caps, allow-lists, kill switch and audit log

Use cases

  1. 01Review portfolio and market data with an AI assistant
  2. 02Preview and validate an order before submission
  3. 03Manage alerts and GTTs through an MCP client

MCP support

The server exposes Zerodha Kite Connect v3 capabilities as tools to MCP clients over stdio or HTTP. It can be configured for read-only use, and write operations are dry-run by default.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

mcp.json · Universal config
{
"mcpServers": {
"techfreakworm-kite-connect": {
"command": "uvx",
"args": [
"kite-connect-mcp"
],
"env": {
"KITE_API_KEY": "<KITE_API_KEY>",
"KITE_API_SECRET": "<KITE_API_SECRET>",
"KITE_MCP_AUTH_MODE": "<KITE_MCP_AUTH_MODE>",
"KITE_TOKEN_FILE": "<KITE_TOKEN_FILE>",
"KITE_MCP_TRADING_MODE": "<KITE_MCP_TRADING_MODE>",
"KITE_MCP_READ_ONLY": "<KITE_MCP_READ_ONLY>"
}
}
}
}
Run commanduvx kite-connect-mcp
Environment variables to set
KITE_API_KEYSecretAPI key of your Kite Connect app (developers.kite.trade)
KITE_API_SECRETSecretAPI secret of your Kite Connect app, needed for the daily login (or use KITE_API_SECRET_FILE)
KITE_MCP_AUTH_MODESecretmanaged (this server owns the daily login) or external (read a token file another program writes)
KITE_TOKEN_FILESecretPath of the access-token file (default: ~/.local/state/kite-connect-mcp/access_token.json)
KITE_MCP_TRADING_MODEdry_run (default: writes are only previewed) or live (writes need a confirmation token)
KITE_MCP_READ_ONLYtrue hides every tool that can change the account

Local servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.

Packages
pypikite-connect-mcp0.1.1

Tools

53

Saved tool list · updated 4d ago

ParameterTypeRequiredDescription
verifybooleanNoalso ask Kite once (GET /user/profile) whether it still accepts the token, and report the account's user id or Kite's error

Details

From the project README

kite-connect-mcp

CI PyPI Licence: MIT

A self-hostable Model Context Protocol server for the Zerodha Kite Connect v3 API. Run it on your own machine or server with your own Kite Connect app, and let an AI assistant read your account, market data and orders. Every write is a dry run by default.

Status: beta (0.x). Every Kite Connect v3 capability is covered and tested offline against a fake Kite. The read tools, the margin and charge calculators and the ticker have also been checked against the live API; live writes (orders, GTTs, alerts, mutual funds) have not, so expect rough edges there and report them. Changes are listed in CHANGELOG.md. This project is not affiliated with or endorsed by Zerodha.

Why another Kite MCP server?

Zerodha publishes an official server (zerodha/kite-mcp-server) and hosts it at mcp.kite.trade. This project differs in three ways:

  • It covers the whole paid Kite Connect API. That includes alerts, GTTs, order and basket margins, charges, position conversion, holdings authorisation, auctions, mutual funds and the WebSocket ticker.
  • It's built for one person self-hosting. The daily token is persisted and shared by every client you connect.
  • It has a safety layer for orders: dry runs, two-step confirmation, value caps, allow-lists, a kill switch, an audit log and rate limiting.

The full evaluation and reasoning are in docs/DECISION.md.

What you need

  • A Zerodha account, and a Kite Connect app from the developer console.
    • Connect (paid, ₹500/month per API key) includes live quotes, the WebSocket and historical candles.
    • Personal (free) covers orders, GTTs, alerts, portfolio and margins, but no market data. Market-data tools return a clear "needs the paid plan" error.
  • Python 3.11+ (or Docker).
  • To place live orders: the order requests must come from a static IP registered in the developer console. This is a SEBI/NSE rule, in force since April 2026. Reads work from any IP.

The developer-console steps are in docs/KITE_SETUP.md: creating the app, the API key and secret, the redirect and postback URLs, the subscription, and the static IP.

Quick start (local, stdio)

# install
uv tool install kite-connect-mcp        # or: pipx install kite-connect-mcp  /  pip install kite-connect-mcp
# or run without installing:  uvx kite-connect-mcp --help
# or the image:               docker pull ghcr.io/techfreakworm/kite-connect-mcp

# configure: credentials from env vars or files (mode 600)
export KITE_API_KEY=your_api_key
export KITE_API_SECRET_FILE=~/.config/kite-connect-mcp/api_secret

kite-connect-mcp check-config   # validates settings, never prints credentials
kite-connect-mcp login --listen # prints the Kite login link and waits for the redirect (managed mode)
kite-connect-mcp status         # session and safety status

For login --listen, register http://127.0.0.1:8765/callback as your app's redirect URL. Every setting is listed in docs/CONFIGURATION.md.

Add it to an MCP client, e.g. Claude Desktop (claude_desktop_config.json) or any client that launches stdio servers:

{
  "mcpServers": {
    "kite": {
      "command": "kite-connect-mcp",
      "args": ["serve"],
      "env": { "KITE_API_KEY": "your_api_key", "KITE_API_SECRET_FILE": "/home/you/.config/kite-connect-mcp/api_secret" }
    }
  }
}

To run it on a server over HTTP, see docs/SELF_HOSTING.md. It covers systemd and Docker, a reverse proxy with TLS, bearer-token auth, and the login callback.

The safety model, in short

GuardDefault
Every write tool takes dry_runtrue: validate, run the checks, preview the exact request and Kite's margin calculation; send nothing
Live trading switchoff (KITE_MCP_TRADING_MODE=dry_run)
Confirmationa live write needs the single-use confirmation_token from a dry run of the identical call (or, if you opt in, an explicitly allow-listed instrument)
Value caps₹10,000 per order, ₹50,000 per day, 20 live orders per day
Allow-listsinstruments, exchanges, products, varieties, order types, sides (all optional)
Read-only modeKITE_MCP_READ_ONLY=true hides every write tool, and the HTTP client refuses write endpoints too
Kill switcha file; a tool can switch it on, only a person can switch it off
Audit logevery dry run, refusal, live send and result, as JSON lines; the intent is fsynced before sending
Rate limitskept under Kite's 1/s (quotes), 3/s (historical), 10/s (orders and the rest), 400 orders/min
Retriesreads back off and retry; writes are never retried. A timeout is reported as an unknown outcome: check the order book before trying again

Details: docs/SAFETY.md.

Sessions and the daily login

Kite access tokens expire every day at about 06:00 IST. A request token is single-use. Logging in again with the same API key invalidates the previous access token. So one app should have exactly one program that owns the daily login.

  • managed mode (default). This server owns the login. The login tool (or kite-connect-mcp login) gives you the link. Kite redirects to this server's /callback, which checks the one-time state, exchanges the request_token, and stores the token with mode 600.
  • external mode. Another program owns the login and writes a token file. This server only reads that file, picks up the new token every morning without a restart, and never writes, deletes or logs out that token.

Tools

58 tools in 11 toolsets (53 by default: the mutual-fund writes are opt-in). Tool names follow the official server where one exists, so prompts carry over. Pick toolsets with KITE_MCP_TOOLSETS and hide single tools with KITE_MCP_EXCLUDED_TOOLS; kite-connect-mcp tools lists what your configuration exposes.

ToolsetTools
sessionsession_status, login, complete_login, logout
accountget_profile, get_margins
ordersplace_order, modify_order, cancel_order, get_orders, get_order_history, get_trades, get_order_trades
gttget_gtts, get_gtt, place_gtt_order, modify_gtt_order, delete_gtt_order
alertsget_alerts, get_alert, get_alert_history, create_alert (simple and ATO), modify_alert, delete_alerts
portfolioget_holdings, get_positions, get_auctions, convert_position, authorise_holdings
marketget_quotes, get_ohlc, get_ltp, get_historical_data, search_instruments, get_instrument, get_instruments_status, refresh_instruments
marginsget_order_margins, get_basket_margins, get_order_charges
mfget_mf_holdings, get_mf_orders, get_mf_sips, search_mf_instruments; opt-in: place_mf_order, cancel_mf_order, place_mf_sip, modify_mf_sip, cancel_mf_sip
tickerticker_subscribe, ticker_unsubscribe, ticker_get_ticks, ticker_status, ticker_order_updates, ticker_stop
safetyget_safety_status, get_audit_log, engage_kill_switch

Every argument of every tool is in docs/TOOLS.md.

Development

git clone https://github.com/techfreakworm/kite-connect-mcp && cd kite-connect-mcp
python -m venv .venv && .venv/bin/pip install -e ".[dev]"
.venv/bin/pytest            # never touches the network: Kite is faked in-process
.venv/bin/ruff check . && .venv/bin/ruff format --check .

See CONTRIBUTING.md. Security reports: SECURITY.md.

Disclaimer

Trading involves risk. This software can place real orders on a real brokerage account when you configure it to. AI systems are unpredictable: keep the caps low, use dry runs, and review what you confirm. The software is provided "as is", without warranty of any kind (see LICENSE). You are responsible for complying with Zerodha's terms, SEBI's rules (including the static-IP and algo rules) and exchange data policies. Kite Connect data may not be redistributed.

Licence

MIT © 2026 Mayank Gupta

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.