Skip to content

Open Agent Search

Open Agent Search
The Eyes of the AI Agent Ecosystem

Privacy-first, unfiltered search for every AI agent. 8 search tools accessible via MCP and REST API — powered by DuckDuckGo.

License: MIT Python 3.12+ FastAPI MCP Compatible
Get Started API Reference
uvx --from open-agent-search oas-mcp


Zero-install — runs directly from PyPI, no clone needed

8
Search Tools
2
Transports
7+
MCP Clients
<100MB
Docker Image

Works Everywhere

Plug Open Agent Search into any MCP-compatible client — or call the REST API directly.

Claude Desktop Claude Code Cursor VS Code Windsurf OpenClaw Any MCP Client

See all MCP setup guides

  • Install in 30 Seconds

    Zero-install with uvx — wire into any MCP client instantly. No clone, no pip install.

    Getting started

  • Full REST API

    Search text, images, videos, news, books — all from one API with interactive Swagger docs.

    API reference

  • MCP Server

    8 tools for Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, OpenClaw & more.

    MCP setup guides

  • One-Click Deploy

    Deploy to Vercel or Docker in seconds. Production-ready out of the box.

    Deploy with Vercel

    Deployment guide

  • Privacy First

    Powered by DuckDuckGo. No tracking, no API keys required, no data collection. SSRF protection built-in.

    Learn more

  • OpenClaw Compatible

    Works as an OpenClaw skill — give your lobster assistant the power of search.

    OpenClaw guide

We are building the eyes of the AI. Independent, agent-neutral access to the world's knowledge.

Feature Description
Dual-mode MCP HTTP transport (/ai/mcp) for deployed servers + stdio for local tools
8 search tools Web, images, videos, news, books, unified search, URL content extraction
Agent-neutral Not bound to any single framework — supports them all
Rate limiting Built-in per-IP rate limiting with configurable thresholds
SSRF protection URL validator blocks private/internal network requests
Zero-install Run via uvx — no clone, no pip install needed
Docker ready Multi-stage build, < 100 MB image, health checks included
Vercel deploy One-click deploy button for serverless hosting
OpenClaw native Works as an OpenClaw skill via stdio MCP transport

Quick Example

{
  "mcpServers": {
    "oas": {
      "command": "uvx",
      "args": ["--from", "open-agent-search", "oas-mcp"]
    }
  }
}

curl "http://localhost:8000/api/search/text?q=python+programming&max_results=5"

# Create the skill directory
mkdir -p ~/.openclaw/workspace/skills/open-agent-search

# Create SKILL.md (see OpenClaw guide for full content)
cat > ~/.openclaw/workspace/skills/open-agent-search/SKILL.md << 'EOF'
---
name: open-agent-search
description: Search the web via Open Agent Search MCP. Privacy-first, powered by DuckDuckGo.
metadata: {"openclaw": {"requires": {"bins": ["uvx"]}, "emoji": "🔍"}}
---
# Open Agent Search
Use `uvx --from open-agent-search oas-mcp` to search the web.
EOF

docker compose up -d
# API → http://localhost:8000
# MCP → http://localhost:8000/ai/mcp

claude mcp add oas -- uvx --from open-agent-search oas-mcp