for ai agents

Search that actually works — free as in freedom, free as in free.

Argand is built for AI agents. No tracking, no behavioral profiling, no API key required to get started. Drop one config block in your agent and ship.

30-second quick start — Claude Desktop

Paste this into your claude_desktop_config.json, restart Claude Desktop, and Argand is live as a search tool.

claude_desktop_config.json json
{
  "mcpServers": {
    "argand": {
      "command": "argand-mcp",
      "args": []
    }
  }
}

Build the binary first: see the Claude Desktop quick start for the one-line install.

Pick your agent

Claude Desktop

→

Drop one JSON block into your config. Restart. Argand is wired in.

Cursor

→

In-app MCP settings or ~/.cursor/mcp_servers.json — both shown.

Cline

→

VS Code extension MCP settings, JSON-shape included.

Custom agent

→

Spawn argand-mcp as a subprocess and speak MCP stdio yourself.

Or roll your own

Argand exposes three orthogonal surfaces. Pick the one your stack already speaks. They all hit the same backend, with the same anonymous-first defaults.

REST
/api/search, /api/lucky, /api/code/search.
gRPC
argand.agent_api.v1 on :50100 (self-hosters).
MCP stdio
search, fetch_url, cite, package_lookup.
Python SDK
sync + async, requests / httpx.
TypeScript SDK
ESM-first, native fetch, .d.ts.
Rust SDK
tokio + reqwest, rustls.

Why bother

  • Anonymous by default
    Send queries without an API key. No DID, no Solid Pod, no sign-up. Higher quotas with an opaque X-Argand-Key header when you need them.
  • Self-hostable
    Every SDK takes a base_url. Run Argand on your own box, point the SDK at it, ship.
  • No tracking, ever
    IPs are scrambled at the gateway before any backend touches them. Query logs are 1-hour TTL for anonymous traffic. Nothing to subpoena.