Add Flixly as a connector in Claude, ChatGPT, Cursor or any MCP client. Ask for a video and it renders with Sora, Veo, Kling, Seedance, Flux, ElevenLabs and 95+ other models — on your Flixly credits.
https://www.flixly.ai/api/mcpPaste this into your client's connector settings. You'll be asked to sign in once.
Pick where you want to use it.
Seven tools rather than one per model — so your client's tool list stays short and new Flixly models show up automatically.
list_modelsBrowse the catalog — model ids, credit cost, supported durations, resolutions, aspect ratios and capabilities.
generate_imageText-to-image, or edit and restyle an existing image by passing image_url. The result is shown inline in the chat.
generate_videoText-to-video, or animate a still with image_url. Returns a task id to poll when a render runs long.
generate_audioText-to-speech for narration and voiceover, or a full music track from a description.
run_modelAny model with raw parameters — lip sync, upscaling, background removal, video translation, first-to-last-frame.
get_generationCollect the result of a generation that was still rendering. Free to call.
get_accountCredit balance, plan, rate limits and usage totals.
Try asking:
Same pricing
MCP generations cost exactly what the REST API and dashboard cost. No surcharge.
Auto refunds
A failed generation returns its credits automatically — you are never charged for a broken render.
Your plan's limits
Rate limits and parallel job limits follow your subscription, shared with the REST API.
Do I need an API key?
No. Claude, ChatGPT, Cursor and other remote-MCP clients use a sign-in flow — paste the URL, approve once, done. An API key is only needed for the stdio bridge, which exists for clients that can't reach a remote server.
How do I disconnect?
Approving creates an API key on your account named after the app, e.g. "Claude (MCP · a1b2c3)". Revoke it under Dashboard → Settings → API keys and the connection stops working immediately.
What does it cost?
The same credits as the rest of Flixly — nothing extra for MCP. Ask the model to check list_models before generating and it will tell you the cost first. Failed generations are refunded automatically.
Why did my video come back as a task id?
Video takes 30-300 seconds, which is longer than a single tool call should block. The job is already submitted and paid for; the model calls get_generation with that id to collect it. Ask it to check again rather than regenerating.
Can I use my own images as input?
Yes, as long as they're at a publicly reachable URL. Local file paths can't be read by a remote server. Upload to Flixly, or any public host, and pass the URL as image_url / video_url / audio_url.
Which models are available?
All 100+ media models in the Flixly catalog, including Sora 2, Veo 3.1, Kling 3.0, Seedance 2, Runway, Luma, Hailuo, Flux, Nano Banana, Ideogram, Imagen, Seedream, ElevenLabs and Suno. New models appear the moment they go live — there's nothing to update.