Deep Research MCP Server

🚀 Open Source Deep Research MCP Server

An AI-based research assistant capable of conducting in-depth, iterative research on any topic. It combines search engines, web scraping, and AI to explore topics and generate comprehensive reports. Available as a Model Context Protocol (MCP) tool or an independent CLI.

🚀 Quick Start

Installation Steps

1. Clone and install:

git clone https://github.com/Ozamatash/deep-research
cd deep-research
npm install

2. Set environment variables in .env.local:

# Copy the example environment file
cp .env.example .env.local

3. Build the server:

# Build the server
npm run build

4. Run the CLI version:

npm run start "Your research query"

5. Test the MCP server with Claude Desktop: Follow the guide below to add the server to Claude Desktop (the MCP version does not ask follow-up questions for now): https://modelcontextprotocol.io/quickstart/server

✨ Features

• Conduct in-depth, iterative research by generating targeted search queries.
• Control the research scope using depth (how deep) and breadth (how wide) parameters.
• Evaluate the reliability of sources through detailed scoring (0 - 1) and inference.
• Prioritize high-reliability sources (≥0.7) and verify less reliable information.
• Generate follow-up questions to better understand research requirements.
• Generate detailed Markdown reports containing findings, sources, and reliability assessments.
• Can be used as a Model Context Protocol (MCP) tool for AI agents.
• The current MCP version does not ask follow-up questions.

🔧 Technical Details

How it Works

flowchart TB
    subgraph 输入
        Q[User Query]
        B[Breadth Parameter]
        D[Depth Parameter]
        FQ[Feedback Question]
    end

    subgraph 深度研究[In-depth Research]
        direction TB
        SQ[Generate SERP Query]
        SR[Search]
        RE[Source Reliability Evaluation]
        PR[Process Results]
    end

    subgraph 结果[Research Output]
        direction TB
        L((What is Learned & 
        Reliability Score))
        SM((Source Metadata))
        ND((Next Steps: 
        Pre - set Goals, 
        New Questions))
    end

    %% Main Process
    Q & FQ --> CQ[Combined Query]
    CQ & B & D --> SQ
    SQ --> SR
    SR --> RE
    RE --> PR

    %% Result Process
    PR --> L
    PR --> SM
    PR --> ND

    %% Depth Decision and Recursion
    L & ND --> DP{Depth > 0?}
    DP -->|Yes| SQ
    
    %% Final Output
    DP -->|No| MR[Markdown Report]

    %% Formatting
    classDef 输入 fill:#7bed9f,stroke:#2ed573,color:black
    classDef 过程 fill:#70a1ff,stroke:#1e90ff,color:black
    classDef 输出 fill:#ff4757,stroke:#ff6b81,color:black
    classDef 结果 fill:#a8e6cf,stroke:#3b7a57,color:black,width:150px,height:150px

    class Q,B,D,FQ 输入
    class SQ,SR,RE,PR 过程
    class MR 输出
    class L,SM,ND 结果

Advanced Configuration

Run Fire Locally

# Install dependencies
pip install fire
# Start the service
fire.Fire(DeepResearch.start)

Observability Tracking (Optional)

Add the following to .env.local:

# Set the observability key
export LANGUAGESPIPELINE_API_KEY=your_api_key_here