Gate97
Sign up

Connecting to an AI

There's no dedicated Claude, OpenAI, or other AI integration in the built-in list — and you don't need one. AI providers expose a plain HTTP API, so the same Custom App mechanism used for any app Gate97 doesn't have built in works here too. Define the AI provider once as a Custom App, then use it as an action step like anything else, referencing its output in the next step with the same field-mapping syntax covered in Robots, Triggers & Actions.

Worked example: auto-write and publish a WordPress post with Claude

A content calendar in Google Sheets supplies a topic; Claude writes a short post about it; WordPress publishes the result. Three steps, no code.

1. Define Claude as a Custom App

From Custom Apps → New:

  • Name: Claude
  • Base URL: https://api.anthropic.com
  • Auth type: API key, header name x-api-key
  • Default request/response format: JSON

Then connect it from the Connections page, pasting a real Anthropic API key. WordPress needs no equivalent setup — it's already a built-in integration (see Custom apps for why it still behaves like one under the hood). Just click Connect on its card and enter your site URL and a WordPress Application Password.

2. Build the robot

Trigger — Google Sheets' New Row Added, on a sheet with one column, Topic. Any real trigger works here; this is simply a believable source of new topics.

Action 1 — Claude. Method POST, path /v1/messages, with an extra header Claude's API requires beyond the API key itself:

{"anthropic-version": "2023-06-01"}

and a templated body:

{
  "model": "claude-sonnet-5",
  "max_tokens": 800,
  "messages": [
    { "role": "user", "content": "Write a short blog post about {{trigger.Topic}}" }
  ]
}

Action 2 — WordPress. Method POST, path /wp-json/wp/v2/posts, body:

{
  "title": "{{trigger.Topic}}",
  "content": "{{step1.content[0].text}}",
  "status": "publish"
}

{{step1.content[0].text}} is Claude's reply — step1 addresses the first action step's output by position, and Claude's Messages API returns its answer as content[0].text in the response body, so this pulls the generated post text straight into WordPress with no intermediate step needed.

Put the robot to work. Add a row to the sheet, and within a minute the poller picks it up, Claude drafts a post, and WordPress publishes it — check that robot's Run History to see each step's real input and output.

Using a different AI provider

The pattern is identical for OpenAI or any other AI API — only the Custom App's base URL, auth header name, and the request/response shape in the action step change. Everything else (the trigger, the field mapping into and out of the AI step, putting the robot to work, run history) stays the same.