@kagan-sh/opensearch is an evidence-backed search tool for AI coding agents. It searches the live web and public code in parallel, then returns structured JSON your agent can act on.

Works as a Claude Code plugin, a standalone MCP server, and an OpenCode plugin.

Install

Claude Code (plugin)

Install from the plugin directory for one-click setup with built-in skills and search command:

/plugin install opensearch

Or install from a local checkout for development:

claude --plugin-dir ./plugin

The plugin bundles the MCP server, a SKILL.md (so Claude knows when to search automatically), and a /opensearch:search command.

Claude Code (MCP — standalone)

If you prefer a standalone MCP server (also works with Cursor, Windsurf, etc.):

claude mcp add opensearch -- npx -y @kagan-sh/opensearch

To enable web search via SearXNG:

claude mcp add opensearch \
  -e OPENSEARCH_WEB_URL=http://localhost:8080 \
  -- npx -y @kagan-sh/opensearch

Or add it to .mcp.json in your project root:

{
  "mcpServers": {
    "opensearch": {
      "command": "npx",
      "args": ["-y", "@kagan-sh/opensearch"],
      "env": {
        "OPENSEARCH_WEB_URL": "http://localhost:8080"
      }
    }
  }
}

See the Claude Code install guide for full details.

OpenCode (plugin)

Add the plugin to opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["@kagan-sh/opensearch"]
}

OpenCode installs npm plugins automatically at startup.

Full docs: kagan-sh.github.io/opensearch. Web search uses a self-hosted SearXNG instance; setup lives in the docs.

License

MIT