Deep Research MCP Server exposes an MCP-compatible interface that orchestrates a Gemini 2.5 Flash-powered deep research pipeline. It supports iterative query refinement, SERP-style query generation, semantic content splitting, batched model calls with tooling (Google Search Grounding, Code Execution, Functions), and produces a structured Markdown report as the final output. The server is designed to be integrated with MCP-aware clients and the Inspector tool, enabling researchers to embed the agent into larger automation workflows. To use it, start the MCP server locally and connect an MCP client to the provided entry point; then issue tasks that leverage the deep research workflow to perform iterative investigations, extract learnings, and generate professional reports. Features like depth/breadth controls and token-aware chunking help tailor exploration and outputs to your research scope. Tools can be toggled via environment flags to enable or disable specific capabilities (e.g., search grounding or code execution) during the research cycles.

Prerequisites:

• Node.js v22.x installed
• Git installed
• Access to Gemini 2.5 Flux API (Gemini API key)
• Optional: Google Search Grounding API key if you plan to use the SERP tool

1. Clone the repository git clone https://github.com/ssdeanx-deep-research-mcp-server.git cd ssdeanx-deep-research-mcp-server

2. Install dependencies npm install

3. Configure environment

   • Create a copy of the example env and populate keys cp .env.example .env

     Edit .env to add GEMINI_API_KEY, GOOGLE_API_KEY, and GOOCLE_CSE_ID if needed

4. Build (if a build step exists) and start the MCP server npm run build # if a build script exists npm start # or run the MCP server directly via Node (see mcp_config for entry)

5. Verify the server is running Check the console for the MCP server startup message and accessible endpoints

Notes:

• If you run directly from TypeScript sources, you may need ts-node or a build step to transpile to JavaScript first.
• Ensure your environment variables are securely stored and not committed to version control.

Tips and common considerations:

• Environment variables: GEMINI_API_KEY is mandatory for Gemini-based reasoning. GOOGLE_API_KEY and GOOCLE_CSE_ID (if using Google Search Grounding) are optional depending on tool usage.
• Tools: You can enable Google Search Grounding, Code Execution, and Functions via flags or environment settings. If a tool is disabled, the corresponding capability will be unavailable in the research pipeline.
• Output: The server emphasizes deterministic, Zod-validated JSON outputs and structured Markdown reports. If outputs appear inconsistent, check batching settings, token limits, and cache configurations.
• MCP integration: Designed to work with MCP-aware clients and Inspector for testing; ensure your MCP client targets the server entry (src/mcp-server.ts) or the compiled entry equivalent in your deployment.
• Performance: The pipeline uses concurrency-limited batching and LRU caches. Adjust concurrency or cache sizes if you run into rate limits or memory constraints.