Scour

A lightning-fast search API in Rust that scrapes Brave and Yahoo (Bing) concurrently and returns normalized JSON. Runs as an HTTP server or MCP server for AI agents. No API keys required.

How It Works

HTTP/MCP request → concurrent scrapers → Brave + Yahoo + DuckDuckGo* → deduped results as JSON

Each engine runs with an independent timeout (DuckDuckGo: 2s, Brave/Yahoo: 5s) so a single hung engine doesn't block results. Duplicate URLs across engines are merged into one result listing all contributing engines.

DuckDuckGo is network-blocked from the production server (Azure IP 20.204.244.192 hangs on TCP connect). It always times out but never blocks other engines.

Quick Start

cargo build --release
./target/release/scour
curl "http://localhost:10080/search?q=hello+world"

HTTP API

GET /search?q=<query>

Response:

{
  "query": "rust tokio",
  "results": [
    {
      "title": "Tokio - An asynchronous Rust runtime",
      "url": "https://tokio.rs/",
      "snippet": "Tokio is a library for writing fast...",
      "engine": "Brave"
    },
    {
      "title": "Tokio - The Tokio Guide",
      "url": "https://tokio.rs/docs",
      "snippet": "A practical guide...",
      "engine": "Brave, Yahoo"
    }
  ]
}

The engine field may contain comma-separated names when the same URL is returned by multiple engines.

MCP Server Mode

For AI agents (Claude Desktop, Gemini CLI, Cursor):

./scour --mcp

Configure:

{
  "mcpServers": {
    "scour": {
      "command": "/path/to/scour",
      "args": ["--mcp"]
    }
  }
}

Exposes one tool web_search that takes a query string and returns aggregated results.

Engines

Engine	Status
Brave	✅ Working
Yahoo (Bing)	✅ Working
DuckDuckGo	❌ Network blocked (production) — always times out

Performance

Tested against live API (api.ahur.ir/search):

Metric	Result
Avg latency (5 samples)	~1.2s
Max concurrent users	10 at 100% success
Batch throughput	30 req / ~4s

Build

cargo build --release

Requires: Rust toolchain.

License

CC BY-NC 4.0