deep-search-mcp

MCP Server for deep web & academic search — with proxy support for restricted networks.

Most MCP search tools only wrap a single web search engine. deep-search-mcp gives your AI agent 6 search backends — including three academic databases that most tools completely ignore — and works seamlessly behind corporate proxies with SSL inspection.

What makes it different

| Feature | Other MCP search tools | deep-search-mcp | |---|---|---| | Web search | ✅ | ✅ DuckDuckGo + Exa AI | | Academic papers | ❌ | ✅ arXiv + Semantic Scholar + PubMed | | Chinese search | ❌ | ✅ Bocha AI | | Proxy support | ❌ | ✅ system / custom URL | | SSL inspection bypass | ❌ | ✅ --ignore-ssl | | Free to use | partial | ✅ 4 out of 6 engines need no API key | | Context7 docs | ❌ | ✅ built-in |

Quick start

# No install needed
npx deep-search-mcp --web-search arxiv

# Behind a corporate proxy with SSL inspection
npx deep-search-mcp --proxy system --ignore-ssl --web-search arxiv

# Custom proxy URL
npx deep-search-mcp --proxy http://proxy.example.com:8080 --ignore-ssl

Search engines

Web search

# DuckDuckGo — free, no API key
npx deep-search-mcp --web-search duckduckgo

# Exa AI — AI-ranked results with content snippets
npx deep-search-mcp --web-search exa

# Bocha AI — best for Chinese content
npx deep-search-mcp --web-search bocha --bocha-api-key sk-xxx

Academic search

# arXiv — preprints across CS, physics, math, biology
# Free, no key needed
npx deep-search-mcp --web-search arxiv

# Semantic Scholar — 200M+ papers with citation metrics
# Free; API key optional (raises rate limit)
npx deep-search-mcp --web-search semantic-scholar
npx deep-search-mcp --web-search semantic-scholar --semantic-scholar-api-key <key>

# PubMed — 35M+ biomedical and clinical papers (NCBI)
# Free; API key optional (raises rate limit)
npx deep-search-mcp --web-search pubmed
npx deep-search-mcp --web-search pubmed --pubmed-api-key <key>

Academic search results

Each academic result includes everything you need to evaluate a paper at a glance:

1. **Attention Is All You Need**
   URL: https://arxiv.org/abs/1706.03762
   作者: Ashish Vaswani, Noam Shazeer, Niki Parmar 等 | 发表: 2017-06-12 | PDF: https://arxiv.org/pdf/1706.03762
   We propose a new simple network architecture, the Transformer...

2. **BERT: Pre-training of Deep Bidirectional Transformers**
   URL: https://www.semanticscholar.org/paper/...
   作者: Jacob Devlin, Ming-Wei Chang | 年份: 2019 | 引用: 50000 | DOI: 10.18653/...
   We introduce BERT, designed to pre-train deep bidirectional representations...

Proxy configuration

The proxy options work for all six search engines and the web fetch tool:

# Auto-detect system proxy (reads env vars on Linux/Mac, registry on Windows)
npx deep-search-mcp --proxy system

# Explicit proxy URL
npx deep-search-mcp --proxy http://proxy.example.com:8080

# With SSL certificate bypass (needed when proxy does TLS inspection)
npx deep-search-mcp --proxy system --ignore-ssl

# No proxy
npx deep-search-mcp --proxy none

Diagnose your proxy connection:

npx deep-search-mcp diagnose --proxy system --ignore-ssl

This runs 5 checks: proxy detection → connectivity → web search → web fetch → Context7.

Claude Code integration

Add to your Claude Code MCP config:

{
  "mcpServers": {
    "deep-search": {
      "command": "npx",
      "args": [
        "deep-search-mcp",
        "--proxy", "system",
        "--ignore-ssl",
        "--web-search", "arxiv"
      ]
    }
  }
}

For biomedical research:

{
  "mcpServers": {
    "deep-search": {
      "command": "npx",
      "args": [
        "deep-search-mcp",
        "--proxy", "system",
        "--ignore-ssl",
        "--web-search", "pubmed"
      ],
      "env": {
        "PUBMED_API_KEY": "your-ncbi-key"
      }
    }
  }
}

MCP tools exposed

| Tool | Description | |---|---| | web_search | Search using the configured engine | | web_fetch | Fetch and convert any URL to markdown / text / HTML | | resolve-library-id | Context7: find a library's ID | | query-docs | Context7: retrieve library documentation |

All options

| Option | Description | Default | |---|---|---| | -p, --proxy <proxy> | system | none | http://... | reads env vars | | --web-search <engine> | duckduckgo | exa | bocha | arxiv | semantic-scholar | pubmed | duckduckgo | | --bocha-api-key <key> | Bocha AI API key | BOCHA_API_KEY env | | --semantic-scholar-api-key <key> | Semantic Scholar API key | SEMANTIC_SCHOLAR_API_KEY env | | --pubmed-api-key <key> | PubMed / NCBI API key | PUBMED_API_KEY env | | -t, --timeout <ms> | Request timeout in ms | 30000 | | --ignore-ssl | Bypass SSL certificate verification | false | | --no-context7 | Disable Context7 tool passthrough | enabled by default | | --context7-api-key <key> | Context7 API key (optional) | CONTEXT7_API_KEY env |

Environment variables

HTTPS_PROXY=http://proxy.example.com:8080
BOCHA_API_KEY=sk-xxx
SEMANTIC_SCHOLAR_API_KEY=xxx
PUBMED_API_KEY=xxx
CONTEXT7_API_KEY=ctx7sk_xxx
IGNORE_SSL=true

Search engine comparison

| Engine | Free | Key required | Best for | |---|---|---|---| | DuckDuckGo | ✅ | No | General web search | | Exa AI | ❌ | No (MCP endpoint) | AI-ranked web results | | Bocha AI | ❌ | Yes | Chinese content | | arXiv | ✅ | No | Preprints (CS, physics, math, bio) | | Semantic Scholar | ✅ | Optional | Cross-discipline papers + citations | | PubMed | ✅ | Optional | Biomedical & clinical research |

Development

git clone https://github.com/your-name/deep-search-mcp.git
cd deep-search-mcp
npm install
npm run dev -- --proxy system --ignore-ssl --web-search arxiv
npm test
npm run build

License

MIT