Deep Research MCP Server

Why not generic?

Architecture

graph LR
    A[User: deep_research topic] --> B[search_web DDG HTML]
    B --> C[Parallel Scrape x3-8]
    C --> D[cheerio clean + turndown md]
    D --> E[extract_insights heuristic]
    E --> F[Synthesize Report + Citations]
    F --> G[memory_save + history]
    F --> H[Return to Claude]
    
    I[compare_sources] --> C
    J[fact_check_claim] --> B
    K[memory_search] --> G

Tools (8)

Resources:

• research://memory — all saved findings

• research://history — last 100 actions

• research://stats — cache size, uptime

Prompts:

• deep-dive-research — full research workflow

• fact-check — fact-checker squad

• compare-narratives — bias & comparison table

Quickstart & Installation

⚡ 1-Click Auto Setup for Claude Desktop (Recommended)

Run this single command in your terminal:

npx mcp-deep-research-server --install

That's it! It automatically detects your OS, writes the config to your claude_desktop_config.json, and configures the 8 research tools. Just restart Claude Desktop.

📦 Install via Smithery

To automatically install for Claude Desktop or Cursor via Smithery:

npx -y @smithery/cli install mcp-deep-research-server --client claude

🛠️ Manual / Other Clients Setup

Run via npx:

npx mcp-deep-research-server

Or build from source:

git clone https://github.com/SECRET4422/mcp-deep-research-server.git
cd mcp-deep-research-server
npm install
npm run build

🤖 Agent Skill (Claude Code / Antigravity / Cursor / Agent Frameworks)

This repository includes an open-standard Agent Skill that teaches AI coding assistants when and how to autonomously run deep research workflows.

• Import via skill directory: skills/deep-research/

• Spec compliant with Claude Code, Cursor, and Google Antigravity agents.

Test (smoke)

npm run test:mcp
# or
npm run inspect # opens http://localhost:6274

Manually tested:

[search] Dehradun → 3 results ✓
[deep_research] What is MCP → 3 sources in 2.1s ✓
tools/list → 8 tools ✓

Add to Claude Desktop

Edit config:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "deep-research": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-deep-research-server/build/index.js"]
    }
  }
}

Restart Claude Desktop.

Add to Cursor / Windsurf / VS Code

.cursor/mcp.json or mcp.json:

{
  "mcpServers": {
    "deep-research": {
      "command": "node",
      "args": ["./build/index.js"],
      "cwd": "/path/to/mcp-deep-research-server"
    }
  }
}

Example Prompts

Deep Research:

Use deep_research to research "Best LLM fine-tuning in 2026, depth deep" then compare LoRA vs QLoRA

Fact Check:

Fact check claim: "Bun is faster than Node" using fact_check_claim

Compare:

Compare these 3 URLs about MCP architecture focusing on security: https://modelcontextprotocol.io/docs/getting-started/intro https://www.anthropic.com/news/model-context-protocol https://en.wikipedia.org/wiki/Model_Context_Protocol

See examples/claude-example.md for more.

Data Storage

All in ~/.mcp-deep-research/:

• memory.json — persistent findings

• history.json — audit log (100 max)

• cache/ — reserved

No DB, no external calls except search/scrape.

Pro Features in v1.1.0

• ✅ Logo + pro README + badges

• ✅ GitHub Actions CI (Node 18/20/22) + Release workflow

• ✅ Issue templates, PR template, CONTRIBUTING, SECURITY

• ✅ .editorconfig, smoke test script

• ✅ Optimized package.json for npm publishing

• ✅ CHANGELOG tracked

Roadmap

• Tavily / Brave API fallback if keys present

• PDF parsing via pdf-parse

• YouTube transcript tool

• Vector search on memory (embeddings)

• SSRF protection — private/internal IP blocklist for scrape_page

• Smithery & Glama registry integration

Dev

npm run dev     # tsx watch
npm run build
npm run lint

Guidelines in CONTRIBUTING.md.

License

MIT © Prabhakar Pal — See LICENSE

Built with 🧠 for Dehradun → World. Not a generic MCP.