n8n, Make and Zapier

Queueing answers straight away with an id, and the work happens after, so every one of these tools needs the same shape: call, wait, check, repeat until the state is done or failed. 3 seconds between checks is plenty. Keep the token in the tool's credential store rather than typing it into a node.

n8n

1. HTTP Request node: POST https://api.youtubetotext.ai/v1/api/transcribe, header Authorization: Bearer ytt_<your-token>, JSON body {"url": "..."}
2. Wait node: 3 seconds
3. HTTP Request node: GET https://api.youtubetotext.ai/v1/api/transcription/{{$json.id}}
4. IF node: state is one of "done", "failed", otherwise loop back to the Wait node

Make

1. HTTP module, Make a request: POST https://api.youtubetotext.ai/v1/api/transcribe with the same header and body
2. Sleep module: 3 seconds
3. HTTP module: GET the transcription by id
4. Router: done or failed ends the scenario, anything else repeats

Zapier

1. Webhooks by Zapier, POST: https://api.youtubetotext.ai/v1/api/transcribe with the header and the JSON body
2. Delay by Zapier: 3 seconds
3. Webhooks by Zapier, GET: the transcription by id
4. Filter: continue only when state is done, so the Zap ends on the result

Or skip the wiring

If the thing on the other end is an assistant rather than a workflow, connect the MCP server at https://api.youtubetotext.ai/mcp and ask it for a transcript in words. It signs in through your account, so there is no token to store and no polling to build. How to connect it.