Deep Research MCP Tool for Cursor AI

Overview

This project is a modular, extensible npm package implementing a Deep Research MCP (Model Context Protocol) tool for use with Cursor AI. It enables robust, autonomous research workflows, integrating real-time web search via the Tavilly API and following best practices for MCP tool development.

Key Requirements

• Modular, extensible architecture for easy refinement and feature addition
• Strict adherence to the deep research process (planning, autonomous search, synthesis, reporting)
• Real data only (no simulation), with Tavilly as the sole external API
• Full compatibility with Cursor AI's agent workflows and approval mechanisms

Architecture

• Core MCP Server: Provides stdio/HTTP transport for Cursor AI integration
• Deep Research Orchestrator: Handles query analysis, planning, and workflow
• Tavilly API Integration: Real web search, robust error handling, and retries
• Research Tools: Web search, data extraction, synthesis, reporting, and citation
• Extensibility Points: Easy addition of new tools/resources
• Config/Environment Management: API keys, runtime options

Setup

1. Clone the repository and install dependencies:

   npm install

2. Build the project:

   npm run build

3. Start the MCP server (for Cursor AI stdio integration):

   npm start

4. Configure your Tavilly API key in a .env file:

   TAVILLY_API_KEY=your_api_key_here

Testing & Development Workflow

• Test Runner: This project uses Jest for unit and integration testing.
• Test Files: All test files are located in the tests/ directory and follow the .test.ts naming convention.
• Running Tests:

  npm test

• Adding Tests: Add new test files or cases in the tests/ directory. Use test-driven development (TDD) to ensure extensibility and robustness as you add new features or tools.
• Best Practices:
  • Write tests for every new tool, orchestrator function, or integration.
  • Use mocks and stubs for external APIs and server interactions where needed.
  • Keep tests up to date as the codebase evolves.

Next Steps

• Implement the Deep Research Orchestrator module
• Add Tavilly API integration with robust error handling
• Develop research tools for web search, extraction, synthesis, and reporting
• Write tests for all modules and ensure robust, real-world data handling
• Document extensibility points for future features

For more details, see cursor-mcp.md and dr-process.md in the project root.