Deep Research

Clone the deep-research repo and install all Node.js dependencies using pnpm, npm, or yarn.

git clone https://github.com/u14app/deep-research.git
cd deep-research
pnpm install

Create a .env.local file (or set environment variables) with your chosen AI provider and search provider credentials. MCP_AI_PROVIDER and MCP_SEARCH_PROVIDER select the active providers; MCP_THINKING_MODEL and MCP_TASK_MODEL specify which models to use.

MCP_AI_PROVIDER=google
MCP_SEARCH_PROVIDER=tavily
MCP_THINKING_MODEL=gemini-2.0-flash-thinking-exp
MCP_TASK_MODEL=gemini-2.0-flash-exp
GOOGLE_GENERATIVE_AI_API_KEY=your_gemini_api_key_here
ACCESS_PASSWORD=your_optional_password

Launch the Next.js development server. The MCP endpoint becomes available at /api/mcp (streamable-http) and /api/mcp/sse (SSE).

pnpm dev

Alternatively, pull and run the pre-built Docker image. Pass environment variables with -e flags.

docker run -d --name deep-research -p 3333:3000 \
  -e MCP_AI_PROVIDER=google \
  -e MCP_SEARCH_PROVIDER=tavily \
  -e MCP_THINKING_MODEL=gemini-2.0-flash-thinking-exp \
  -e MCP_TASK_MODEL=gemini-2.0-flash-exp \
  -e GOOGLE_GENERATIVE_AI_API_KEY=your_key \
  xiangfa/deep-research:latest

Add the deep-research server to your MCP client configuration using the streamable-http transport. Set a long timeout (600+ seconds) as research can take up to 2 minutes. If ACCESS_PASSWORD is set, include the Authorization header.

{
  "mcpServers": {
    "deep-research": {
      "url": "http://127.0.0.1:3000/api/mcp",
      "transportType": "streamable-http",
      "timeout": 600
    }
  }
}