Connect an agent to Rindler

The CLI is the shortest path. It signs you in, mints a key tied to your session, and configures Claude Code and Codex for you:

curl https://rindler.ai/cli | sh
rindler login

The CLI is a dashboard tool. It signs in through app.rindler.ai, and the devices and keys it creates appear on your dashboard. See the CLI guide for the full command list.

The install script writes a .mcp.json in your project and leaves any other servers alone:

curl https://rindler.ai/install | sh

By hand, add this to your .mcp.json. No key needed: you are prompted to sign in on first use.

{
  "mcpServers": {
    "rindler": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.rindler.ai/mcp"]
    }
  }
}

Adding custom connections in the Claude Desktop app is still a beta feature and may need enabling in its settings.

The CLI mints a key bound to your sign-in, which is what you want on a laptop: it lapses when your session does, and there is nothing to revoke by hand.

For CI or a server, mint a long-lived key in the dashboard and set RINDLER_API_KEY. It takes precedence over everything else and is never written to disk. To use one in a config file instead, send it as a bearer header:

{
  "mcpServers": {
    "rindler": {
      "url": "https://mcp.rindler.ai/mcp",
      "headers": { "Authorization": "Bearer rindler_live_YOUR_KEY" }
    }
  }
}

Tokens are shown once when you create them. Copy it then; the dashboard cannot show it again.

Every call an agent makes is an HTTP call you can make yourself. List what you can drive:

curl -s https://mcp.rindler.ai/v1/runtime/configs \
  -H "Authorization: Bearer $RINDLER_API_KEY"

The response is an envelope, not a bare array: {"configs": [...]}.

Running an action is two calls. The first returns a job id:

curl -s -X POST https://mcp.rindler.ai/v1/runtime/run \
  -H "Authorization: Bearer $RINDLER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"site": "example.com", "actions": [{"name": "search", "inputs": {"query": "laptops"}}]}'

The second reads the result:

curl -s https://mcp.rindler.ai/v1/runtime/jobs/JOB_ID \
  -H "Authorization: Bearer $RINDLER_API_KEY"

A job can finish and still have retrieved nothing, so status and retrieval.complete answer different questions.

The CLI encodes exactly this: rindler run exits non-zero when retrieval is incomplete, even on a job that finished.

Run rindler doctor first. It checks your config, your key, both agent installs, and whether the server is reachable.

• The full command list is in the CLI guide.
• For what the tools actually do once you are connected, see the docs overview.