Enhanced Browser MCP

A crew of AI agents that handle web research so Claude doesn't have to.

Enhanced Browser MCP delegates research to a specialized crew of AI workers. One tool call triggers parallel searches across 25+ sources, intelligent scraping, and AI synthesis - returning concise findings instead of raw web content.

90%+ Token Reduction

| Without Enhanced Browser | With Enhanced Browser | |--------------------------|----------------------| | Claude searches, gets 10+ URLs | AI crew searches 25+ sources in parallel | | Claude reads each page (5,000+ tokens each) | Jina Reader scrapes pages (free) | | Claude synthesizes manually | Gemini synthesizes findings (free) | | 50,000+ tokens consumed | ~200-1200 tokens returned |

The crew does the heavy lifting. Claude gets the insights.

How It Works

Claude Code
    │
    │ research("your question")
    ▼
┌─────────────────────────────────────────────┐
│         Enhanced Browser MCP                 │
│         (AI Research Crew)                   │
│                                              │
│  ┌──────────────────────────────────────┐   │
│  │     Query Intelligence Layer          │   │
│  │  Detects: code, finance, medical,     │   │
│  │  academic, weather, packages, etc.    │   │
│  └─────────────────┬────────────────────┘   │
│                    │                         │
│    ┌───────────────┼───────────────┐        │
│    ▼               ▼               ▼        │
│ ┌──────┐    ┌───────────┐    ┌──────────┐  │
│ │Search│    │  Scrape   │    │ Synthesize│  │
│ │ Crew │───▶│   Crew    │───▶│   Crew   │  │
│ │(25+) │    │  (Jina)   │    │ (Gemini) │  │
│ └──────┘    └───────────┘    └──────────┘  │
│                                              │
│  Returns: Key findings, details, sources     │
└─────────────────────────────────────────────┘

25+ Search Sources

The crew automatically selects relevant sources based on your query:

General Web (API key needed)

Serper - Google results (2,500 free/month)
Brave - Privacy-focused search (2,000 free/month)
Tavily - AI-optimized search (1,000 free/month)
Exa - Neural/semantic search (1,000 free/month)

Code & Packages (17 free sources)

GitHub - Code and repositories
StackOverflow - Programming Q&A
npm - Node.js packages
PyPI - Python packages
crates.io - Rust packages
Docker Hub - Container images
Libraries.io - Cross-platform packages (API key optional)

Academic & Medical

ArXiv - Scientific preprints (free)
PubMed - Medical literature (free)
Semantic Scholar - Academic papers (free)
Wikipedia - Encyclopedia (free)

Knowledge & Reference

Wolfram Alpha - Computational knowledge (2,000 free/month)
Wikidata - Structured facts (free)
Open Library - Books (free)

Finance

Alpha Vantage - Stocks and forex (25 free/day)
CoinGecko - Cryptocurrency data (free)

Other

OpenWeatherMap - Weather data (1,000 free/day)
Nominatim - Geocoding (free)
HackerNews - Tech discussions (free)
Reddit - Community discussions (API key needed)
NewsAPI - News articles (100 free/day)

Quick Start

1. Clone and Install

git clone https://github.com/bigph00t/enhanced-browser-mcp.git
cd enhanced-browser-mcp
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -e .

2. Get API Keys

Required:

| API | Free Tier | Get Key | |-----|-----------|---------| | Google Gemini | Generous | aistudio.google.com/app/apikey |

Recommended (better results):

| API | Free Tier | Get Key | |-----|-----------|---------| | Serper | 2,500/month | serper.dev | | Brave | 2,000/month | brave.com/search/api |

17 free sources work without any API keys - see .env.example for full list.

3. Configure

cp .env.example .env
# Edit .env with your API keys

4. Add to Claude Code

Add to ~/.claude.json:

{
  "mcpServers": {
    "enhanced-browser": {
      "type": "stdio",
      "command": "/path/to/enhanced-browser-mcp/venv/bin/enhanced-browser",
      "args": [],
      "env": {
        "GOOGLE_API_KEY": "your-gemini-key",
        "SERPER_API_KEY": "your-serper-key",
        "BRAVE_API_KEY": "your-brave-key"
      }
    }
  }
}

5. Restart Claude Code

Usage

Just use research() for everything:

research("your question here")

The crew automatically selects engines based on your query:

| Query Type | Engines Auto-Selected | |------------|----------------------| | Code questions | + GitHub, StackOverflow, npm, PyPI, crates | | Finance questions | + Alpha Vantage, CoinGecko | | Medical questions | + PubMed, Semantic Scholar | | Academic topics | + ArXiv, Wikipedia, Semantic Scholar | | Math/calculations | + Wolfram Alpha | | Weather queries | + OpenWeatherMap | | Location queries | + Nominatim | | Opinions/comparisons | + HackerNews, Reddit |

Depth Options

research("What is HTMX?", depth="quick")           # ~10s, simple facts
research("How to implement rate limiting")          # ~30s, default
research("Rust vs Go for CLI tools", depth="deep")  # ~60s, comprehensive

| Depth | Time | Tokens | Best For | |-------|------|--------|----------| | quick | ~10s | 200-400 | Facts, definitions, "what is X" | | medium | ~30s | 400-800 | How-to, technical docs (DEFAULT) | | deep | ~60s | 800-1500 | Comparisons, thorough research |

Add Context

research("best TTS API", context="for YouTube automation, comparing cost and quality")

Tools Reference

| Tool | Description | |------|-------------| | research | Main tool - the crew searches, scrapes, synthesizes | | research_async | Start research in background | | research_status | Check if background task is done | | research_wait | Wait for background task and get results | | scrape | Read a specific URL via Jina Reader |

Async Research

For deep research while continuing other work:

# Start background research
task = research_async("PostgreSQL indexing best practices", depth="deep")
# Returns: {"task_id": "a1b2c3d4", "status": "started"}

# Continue working...

# Get results when ready
results = research_wait(task_id="a1b2c3d4")

Example Output

## Research Results (medium)
**Query:** FastAPI rate limiting with Redis

## Key Findings
FastAPI doesn't include built-in rate limiting. The standard solution is **SlowAPI**,
a library that wraps the `limits` package and integrates with FastAPI's dependency
injection system.

## Important Details
- SlowAPI decorator syntax: `@limiter.limit("5/minute")`
- For distributed deployments, use Redis backend
- Alternative: Implement at reverse proxy level (nginx, Cloudflare)

## Recommended Links
- [SlowAPI Documentation](https://slowapi.readthedocs.io)
- [FastAPI Production Guide](https://fastapi.tiangolo.com/deployment/)

---
*Engines: serper, brave, stackoverflow | Pages scraped: 3 | Results: 12*

History & Session Awareness

The crew remembers your searches and uses them to provide better, non-repetitive results.

How it works:

Each search saves findings, scraped content, and sources to history/
Related searches are automatically detected via keyword matching
Gemini sees previous findings and focuses on NEW information
Supports up to 1000 historical searches

Progressive research example:

research("what is WebSockets", depth="quick")        # Baseline knowledge
research("WebSocket vs SSE", depth="medium")         # Finds 1 related, avoids repetition
research("real-time architecture patterns", depth="deep")  # Finds 2 related, builds on prior

History is stored locally in history/ and is gitignored by default.

Configuration

Environment Variables

# Required
GOOGLE_API_KEY=your_gemini_key

# Recommended
SERPER_API_KEY=your_serper_key
BRAVE_API_KEY=your_brave_key

# Optional premium
TAVILY_API_KEY=your_tavily_key
EXA_API_KEY=your_exa_key
NEWSAPI_KEY=your_newsapi_key
WOLFRAM_APP_ID=your_wolfram_key
ALPHAVANTAGE_API_KEY=your_alphavantage_key
OPENWEATHER_API_KEY=your_openweather_key
LIBRARIES_API_KEY=your_libraries_key
GITHUB_TOKEN=your_github_token

The crew automatically detects which APIs are configured and only uses those.

Troubleshooting

"No results found"

Check GOOGLE_API_KEY is set (required for synthesis)
Add at least one search API key (Serper, Brave) for better results
Free sources (DuckDuckGo, Wikipedia) work without keys

Slow responses

deep mode takes 45-90 seconds by design
Use quick or medium for faster results

Missing expected sources

Check .env has the API key for that source
The crew only calls configured engines

Development

pip install -e .

# Run server directly
python -m search_crew.server

# Test research function
python -c "
import asyncio
from search_crew.research import research
result = asyncio.run(research('test query', depth='quick'))
print(result)
"

License

MIT

Credits

Jina Reader - Free scraping
Google Gemini - Free AI synthesis
Serper, Brave, Tavily, Exa - Search APIs
Wolfram Alpha, Alpha Vantage, CoinGecko - Specialized data
HackerNews Algolia, ArXiv, Wikipedia - Free sources

MCP Servers