Vercel¶

Deploy Open Agent Search to Vercel with one click — no server management required.

One-Click Deploy¶

Click the button above to:

1. Fork the repository into your GitHub account
2. Create a new Vercel project linked to the fork
3. Deploy automatically

How It Works¶

Vercel uses the @vercel/python builder to serve the FastAPI app as a serverless function. The included vercel.json handles routing:

{
  "builds": [
    {
      "src": "open_agent_search/app.py",
      "use": "@vercel/python"
    }
  ],
  "routes": [
    {
      "src": "/(.*)",
      "dest": "open_agent_search/app.py"
    }
  ]
}

All requests are routed to the FastAPI application, including:

• REST API endpoints (/api/search/*, /api/content/*)
• MCP server (/ai/mcp)
• Interactive docs (/docs, /redoc)

Environment Variables¶

Set environment variables in Vercel → Project → Settings → Environment Variables:

Using the Deployed Instance¶

Once deployed, your instance is available at https://<your-project>.vercel.app:

# REST API
curl "https://your-project.vercel.app/api/search/text?q=hello"

# MCP endpoint
# Point MCP clients to: https://your-project.vercel.app/ai/mcp

Connect MCP Clients to Your Deployment¶

claude mcp add --transport http oas-remote https://your-project.vercel.app/ai/mcp

{
  "mcpServers": {
    "oas-remote": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://your-project.vercel.app/ai/mcp"]
    }
  }
}

Limitations¶

• Vercel serverless functions have a 10 s default timeout (30 s on Pro). Unified search with many results may hit this limit.
• Cold starts add ~1–2 s on the first request after idle time.
• In-memory rate limiting resets per invocation. For persistent rate limits, use a Redis-backed deployment (e.g., Docker).