Give your AI assistant a video editor.

Quick start

1. 1Create a token at Account → Developer. It looks like vcp_live_… and is shown once.
2. 2Paste the config for your client below.
3. 3Ask for what you want, in plain language.

claude mcp add --transport http vibeclipper https://api.vibeclipper.in/v1/mcp \
  --header "Authorization: Bearer vcp_live_your_token_here"

Then, in your assistant:

Clip https://youtube.com/watch?v=… into 8 Shorts with Hormozi captions,
then schedule them across YouTube Shorts and Instagram over the next 7 days.
Show me the plan before anything posts.

How it works

Your assistant sees a set of tools, each with a description written for a model rather than a developer. It chooses which to call and in what order. The normal sequence is:

import_youtube      → a project
generate_clips      → a job (returns immediately; rendering takes minutes)
wait_for_job        → blocks with live progress, or poll get_job
list_clips          → the finished clips, best first
update_clip         → rewrite titles and descriptions
schedule_clips      → spread them across days and platforms

Clipping runs asynchronously, so nothing ever blocks your conversation. Long operations stream progress back as they advance through the pipeline.

What it can do

Permissions

Every token carries a set of scopes, and tools outside those scopes aren't even listed — an assistant that can see a tool will eventually try it, so we hide rather than advertise-and-refuse.

Pick the narrowest one that does the job. A read-only token that can inspect your projects and report status but cannot spend a credit or post anything is a safe thing to leave running unattended.

Errors, rate limits and retries

A failed call comes back as a readable result, not an opaque transport error, so the model can decide what to do next:

{
  "content": [
    {
      "type": "text",
      "text": "Error [payment_required]: You're out of credits…"
    }
  ],
  "isError": true,
  "_meta": {
    "vibeclipper/error": {
      "code": "payment_required",
      "message": "You're out of credits. Buy a credit pack to keep clipping.",
      "retryable": false,
      "actionUrl": "https://vibeclipper.in/pricing"
    }
  }
}

Every token gets 120 calls a minute. Tools that spend credits are tighter — generate_clips is 10 a minute. Throttled calls return rate_limited with a retryAfterMs to sleep on.

Anything that charges you accepts an idempotency_key. Pass one and a retry replays the original response instead of starting — and billing for — a second run. That matters more with agents than with people: they retry a lot.

Built-in prompts

Reusable templates that encode a format's editorial rules plus the tool sequence to run. In Claude Code they show up as slash commands.

Security

• Only a hash of your token is stored. The plaintext exists exactly once, in the response that created it — we cannot recover it, and neither can anyone who reads our database.
• Tokens are accepted from request headers only, never a query string. Query strings end up in access logs, browser history and referrer headers.
• Every call is logged with a secret-scrubbed argument digest. You can see exactly which app did what, and when, on the Developer page.
• Revoke or rotate at any time — it takes effect on the very next call, with no grace period.
• You can disconnect a single app without revoking the token it shares with your other tools.

Versioning

The surface is versioned in the URL (/v1/mcp). Within a version we only add tools and add optional arguments — never remove or rename. A breaking change ships as a new version served alongside the old one, so an integration you set up today keeps working.