Deep Research MCP Server 🚀

What is Deep Research MCP Server?

Deep Research MCP Server is an AI-powered research assistant that performs iterative, deep research on any topic by combining search engines, web scraping, and Gemini large language models. It is available as a Model Context Protocol (MCP) tool for seamless integration with AI agents.

How to use Deep Research MCP Server?

Install Node.js v22.x, set up environment variables (GEMINI_API_KEY, FIRECRAWL_KEY, DATABASE_URL for PostgreSQL), then build with npm run build. To use as an MCP tool, start the server with node --env-file .env.local dist/mcp-server.js. Alternatively, use the CLI directly with npm run start "your research query".

Key features of Deep Research MCP Server

• MCP Integration for use with AI agents
• Iterative research with automatic deepening
• Intelligent query generation using Gemini LLMs
• Configurable depth (1–5) and breadth (1–5)
• Produces detailed markdown reports with sources
• Concurrent search and result processing
• Persistent knowledge storage via PostgreSQL

Use cases of Deep Research MCP Server

• Investigate the latest developments in quantum computing across multiple rounds
• Build a persistent knowledge base that improves over research sessions
• Automate deep-dive literature reviews for technical or academic topics
• Generate comprehensive, cited reports for business intelligence

FAQ from Deep Research MCP Server

What is Deep Research MCP Server and how does it differ from a simple search?

Unlike a single web search, Deep Research MCP Server iteratively refines its research direction: it generates follow‑up search queries based on previous learnings and can dive deeper (up to 5 levels) while keeping a broad scope (up to 5 branches per level).

What are the runtime requirements?

It requires Node.js v22+, a Firecrawl API key for web search and content extraction, a Gemini API key for LLM processing, and a PostgreSQL database (local or remote) for persistent storage.

How does the server store and manage research data?

All research findings and visited URLs are stored in a PostgreSQL database. Previously processed URLs are tracked to avoid duplicates, and the agent can reference past learnings to guide new research directions.

What parameters control the research depth and breadth?

The MCP tool accepts query (string), depth (1–5, how many levels of recursion), breadth (1–5, how many parallel directions per level), and an optional existingLearnings array to build upon prior results.

How can I start the MCP server?

After building the project, run node --env-file .env.local dist/mcp-server.js. The server listens for MCP‑compatible client requests. You can also test it with npx @modelcontextprotocol/inspector node ….