MCP Integration¶
Open Agent Search provides a full Model Context Protocol (MCP) server with 8 search tools. It works in two modes:
| Mode | Command | Transport | Use case |
|---|---|---|---|
| HTTP | open-agent-search | Streamable HTTP | Deployed API — MCP at /ai/mcp |
| stdio | oas-mcp | Standard I/O | Local tool — wire into any MCP client |
Zero-Install¶
No clone, no pip install needed
Just point your MCP client to: uvx --from open-agent-search oas-mcp This uses uvx to run the server directly from PyPI.
Choose Your Client¶
-
Claude Desktop
-
Claude Code
-
Cursor
-
VS Code / Copilot
-
Windsurf
-
OpenClaw
-
Other Clients
Config File Locations¶
| Client | Config file |
|---|---|
| Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) |
%APPDATA%\Claude\claude_desktop_config.json (Windows) | |
| Claude Code | No file — use claude mcp add CLI |
| Cursor | ~/.cursor/mcp.json |
| VS Code | .vscode/mcp.json (per project) |
| Windsurf | ~/.codeium/windsurf/mcp_config.json |
| OpenClaw | ~/.openclaw/openclaw.json or ~/.openclaw/workspace/skills/ |
Available Tools¶
All 8 tools are exposed in both HTTP and stdio modes:
| Tool | Description |
|---|---|
search_web | Text / web search |
search_images | Image search with size & colour filters |
search_videos | Video search with resolution & duration filters |
search_news | News search with time-limit filter |
search_books | Book search |
search_everything | Parallel search across all sources |
fetch_content | Extract content from a single URL |
fetch_multiple_contents | Extract content from multiple URLs (max 10) |
Tool details
See the MCP Tools reference for full parameter docs.