Skip to main content
Base URL: https://mediamonster.app New here? Start with the Quickstart — a free test key and a first call in five minutes. mediamonster.app/developers is the front door in the app, and Status says whether everything is up. Mint a key at mediamonster.app/keys.
API and MCP access comes with every plan, Cyclops included. Credits gate the work, not the key. Key counts and per-minute ceilings are per plan; see Credits & plans.

Authentication

Send the key as a bearer token. Live keys start mm_live_, test keys mm_test_, and both are shown once.
A key does only what its scopes allow, and can be given an expiry date and a list of addresses it may be used from. See Authentication. A test key runs every check a live key does and charges nothing: each run comes back at once with a stand-in file of the right kind. Build against one, then swap in a live key. See Test mode.

Endpoints

The Scope column is what the key has to be minted with — see Scopes. A run started with save_to_vault needs vault:write as well as run. GET /api/v1/skills is the live list of option ids. It doesn’t need a key.

Skills and prices

Same catalogue the Workbench and MCP read. Send option ids on POST /api/v1/run/{skill}. An unknown option is an error. See Skills for the same prices grouped by drawer, and Documents for the PDF tools.

Generate from a prompt

Send a prompt, and pick a model or leave it for the default — each model is billed at its own price. Image and video generation also take a file and start from it: a picture to work from, or the first frame of a clip. Voice, music and sound effects take no file. Model ids and per-model prices are on Generators. avatar-video bills on how long the prompt takes to say. With script set to topic, the prompt is what the video is about instead. The script is written to fit length, the run bills on length, and the script comes back as a text file beside the video.

How requests behave

A skill is charged before the provider is called. A failed run is refunded. The response includes credits_charged.
Download links expire after ten minutes. GET the job or run again for fresh ones.
An option the skill doesn’t have is refused. A typo must not quietly hand back the default look.
Image: png, jpeg, webp (default WebP). Video: mp4, webm (default MP4). Audio and documents are 400 unsupported_media — use the matching skill. Anything else is 400 unsupported_format.
A call waits about 150 seconds, then returns 202 with poll_url. Send wait=false to get a job or run id immediately.
REST and MCP share the same per-key ceiling: 30 / 60 / 300 / 1,000 requests a minute on Cyclops / Minotaur / Kraken / Behemoth. Every key and session on one account shares a second ceiling of twice that, so spreading work over more keys doesn’t buy more throughput. Every response says where you stand in X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset; a 429 adds retry-after. See Rate limits.
X-Request-Id is on every response, and in every error body as request_id. Quote it to support. Send your own X-Request-Id and we use it instead.
dry_run=true on /run/{skill}, /runs and /convert checks everything and answers with the price, step by step, without running. See Dry runs.
Every job and run carries livemode: true for a live key, false for a test key, whose runs also say "test": true.

Idempotency

A request can time out after it has already started a run. Send an Idempotency-Key header with any POST under /api/v1 and retrying it is safe:
  • The first request with a key runs, and its answer is kept for 24 hours.
  • A retry with the same key and the same request — the same method, path and body — gets that answer back, with Idempotent-Replayed: true. Nothing runs twice and nothing is charged twice.
  • The same key on a different request is 422 idempotency_key_reused.
  • A retry while the first is still running is 409 idempotency_in_progress, with retry-after.
  • A 5xx isn’t kept. The key is freed, so the retry runs.
  • A signing secret isn’t kept either. A replayed webhook endpoint create or rotate answers with the endpoint and "secret": null — the first answer was the only one to carry it. Lost it? Rotate again.
A key is 1–255 printable characters; a UUID per request is the easy choice. Keys belong to the credential that sent them — one API key, your signed-in session, or one connected app — so the same Idempotency-Key from another API key is a new request. MCP’s POST /api/mcp doesn’t take one: a single request there carries many calls.

Versions and changes

This is v1. It only grows: new endpoints, new fields and new codes arrive without notice, and nothing you rely on is removed or renamed. See Versioning and the Changelog. Agents should start at MCP.