kite-mcp

Converts conversational natural language requests into executable stock orders on Zerodha Kite by parsing intent (buy/sell), quantity, price, and order type from user messages and routing them through the Kite API. Uses MCP tool schema to map user utterances to structured order parameters with validation before submission.

Manages Zerodha Kite login sessions automatically by intercepting TOTP (Time-based One-Time Password) requirements, generating codes from a stored secret, and maintaining authenticated session state across multiple API calls. Eliminates manual re-authentication by storing and refreshing access tokens server-side.

Interprets conversational trading requests and maps them to specific trading actions (buy, sell, modify, cancel) with appropriate parameters. Uses LLM context to disambiguate intent, infer missing parameters from conversation history, and validate trading logic before execution.

Coordinates execution of multi-leg trading strategies (e.g., buy stock + set stop loss + set profit target) by sequencing dependent orders and handling partial failures. Executes primary order first, then conditionally executes secondary orders based on primary execution status.

Maintains conversation context across multiple trading requests, remembering previously mentioned symbols, quantities, and trading preferences. Enables anaphoric references (e.g., 'sell it' referring to previously mentioned stock) and reduces repetition in multi-turn trading conversations.

Fetches current holdings, quantities, entry prices, and P&L metrics from Zerodha Kite API and structures them for natural language querying. Aggregates position data across segments (equity, derivatives) and calculates derived metrics like total portfolio value, unrealized gains, and position-level returns.

Fetches real-time stock quotes, OHLC data, and market depth information from Zerodha Kite for specified symbols. Queries the Kite quote API with symbol lookup and returns structured price, volume, and bid-ask spread data suitable for order decision-making.

Creates Good-Till-Triggered (GTT) conditional orders on Zerodha Kite that automatically execute when specified price conditions are met. Accepts natural language trigger conditions (e.g., 'buy if price drops to 2000') and converts them into GTT rule objects with order parameters.

Fetches historical order records and executed trades from Zerodha Kite, including order status, execution price, quantity, and timestamp. Structures trade history for analysis and enables LLM agents to understand past trading activity and performance.

Retrieves real-time account balance, available margin, and margin utilization from Zerodha Kite. Calculates available buying power and alerts when margin usage exceeds safe thresholds, enabling risk-aware order placement.

Searches Zerodha's instrument database to resolve natural language stock names (e.g., 'Tata Consultancy Services') to valid trading symbols (e.g., 'NSE:TCS'). Uses fuzzy matching and exchange-aware lookup to handle ambiguous or partial symbol references.

Cancels pending orders or modifies order parameters (quantity, price, order type) before execution on Zerodha Kite. Accepts natural language cancellation requests (e.g., 'cancel my TCS buy order') and maps them to specific order IDs for API submission.

Exposes all trading capabilities as MCP-compliant tools with standardized JSON schemas, enabling LLM clients to discover available operations, understand required parameters, and invoke tools through the Model Context Protocol. Implements tool metadata including descriptions, parameter types, and return schemas.