Hosted captions MCP for Cursor: burned-in subtitles + SRT/VTT in three tool calls, no sign-up

I’m the maker of AwesomeVideo, a hosted captioning API with a remote MCP server. Sharing it here because the whole first run happens inside Cursor’s agent, without opening a website or making an account.

Add it to .cursor/mcp.json:

{
  "mcpServers": {
    "awesomevideo": {
      "url": "https://api.awesomevideoeditor.com/mcp?via=cursorforum"
    }
  }
}

Then ask the agent to caption a video. Without a key, it can do this:

  1. create_key {} (free) returns a token once plus a small proof-of-work challenge.
  2. Put Authorization: Bearer <token> in the server’s headers, reconnect, and call create_key {challenge, nonce} to claim 10 free welcome credits (the PoW takes about 1-3 s; it’s there so one network can’t farm the trial).
  3. caption_video {"sourceUrl": "https://awesomevideoeditor.com/samples/cc0-interview-es.mp4", "preset": "hormozi-pop"}. That 8-second sample costs 2 credits and returns the MP4 with burned-in captions, plus SRT, VTT and a JSON transcript.

estimate and get_catalog are free without a key, so the agent can check the price before it spends anything. Pricing is 1 credit = 1 cent, 5 credits per audio minute for captions. If the balance runs short, the call returns 402 with a checkout link for a human instead of failing halfway, and nothing is charged.

What I’d love feedback on from Cursor users: does the agent follow the “estimate first, then spend” flow on its own, and is the key-in-headers step clear enough, or would you rather use the OAuth endpoint (/mcp/oauth)?

Docs: Video Captions MCP Server for Claude, Cursor and ChatGPT | Awesome Video Editor