list_models
Inputs: None
Video and image catalog with prices, supported durations and formats. Image generation is not an MCP action yet.
Hyference / Developers
Connect your assistant to Hyference
A complete guide to connecting your account, checking costs, generating video and retrieving results through MCP.
MCP connection
MCP lets an external AI assistant call Hyference tools on your behalf. OAuth connects your own account without copying passwords, session tokens or API keys. Your normal account permissions still apply.
https://hyference.lovable.app/mcpRelease note: this guide describes the prepared seven-tool integration. The app must be republished after MCP changes before this public URL serves that revision. A complete connection from an external assistant has not yet been verified; no client-specific compatibility is guaranteed.
1. Connect your account
- Use a client supporting remote MCP over Streamable HTTP and OAuth with dynamic client registration. Availability can depend on the client's plan or workspace administrator.
- Add a custom or remote MCP connection. Name it Hyference, paste the URL above and choose OAuth if an authentication choice is offered.
- Follow the sign-in window to Hyference. Sign in or create your account, then return to the connection request.
- Check the requesting client's name and your account. Approve only a client you trust: access includes starting paid video jobs, not just reading data. Cancel if you do not want to grant access.
- Return to your assistant and refresh its tools if necessary. Ask it to list Hyference models and your balance. This checks the connection without generating a video.
2. What your assistant can do
Available tools
Compare models, quote video prices, read your balance, start text-to-video renders, follow progress, retrieve downloads and list your renders, projects, characters and voices.
Not available through MCP
Image generation, reference-photo uploads, motion transfer, character swap, voice generation or cloning, lip-sync, prompt enhancement, project editing and attaching renders to a project. The current tools cannot delete, top up or publish to Explore.
Listing a saved character or voice does not apply it to a render. Use Studio for existing visual project workflows.
3. Your first video, safely
Start with a read-only request:
“List my Hyference video models and balance. Quote a supported short vertical video of a glass perfume bottle with a slow camera move. Show me the model, duration and total price. Do not generate until I approve.”
- The assistant calls
list_models,get_balanceandquote_render. Select the exact ID and a supported duration from the returned catalog, not a guessed provider name. - After you approve the quoted cost, it calls
create_videowith your prompt, model, duration and ratio. It should generate and save a UUIDrequest_keybefore submission. - Keep the returned render ID. Check
get_renderperiodically for status and billing rather than creating another render. Closing the assistant does not cancel an accepted job. - When completed, open or download the returned video URL. You can also find the result in your Hyference Reel. Refresh an expired link with another
get_rendercall.
Retry rule: reuse the same request_key for retries of the same video. Without a supplied key, each new create_video call gets a new key and may create another paid job. Change the key only for an intentionally new render.
MCP tool reference
quote_render
Inputs: model, duration
Current video price in USD. Use a video model ID returned by list_models; no funds are reserved.
get_balance
Inputs: None
Your current credit balance in USD.
create_video
Inputs: prompt, model, duration, ratio; optional camera, request_key
Starts a paid text-to-video job; returns renderId and requestKey. Prompt: 1–1,000 characters. Ratio: 16:9 or 9:16. Duration must match the selected model. request_key is a UUID.
get_render
Inputs: id (render UUID)
Status, progress, error, cost and billing state. Completed videos include a download URL valid for one hour; call again for a fresh link.
list_my_renders
Inputs: Optional limit (1–50; default 10)
Your most recent video renders, newest first, with status and cost.
list_library
Inputs: kind: projects, characters or voices
Up to 100 names and IDs from your chosen private library. Does not create or edit library entries.
These are MCP tool names, not REST endpoints. All seven tools require the OAuth connection. Video prices and supported settings come from the same catalog as Studio.
Privacy, costs & control
- The client acts as your account. Other users' private renders and libraries are not accessible. The assistant service receives the data returned by the tools you let it call.
- Video creation uses Studio's server-validated pricing and credit reservation. Accepted jobs reserve credit, successful jobs capture it, and settled failures refund the reservation. An uncertain provider outcome may remain reserved while it is reconciled; check the returned billing state.
- Ask for a quote and require your approval before every paid job. This is an assistant instruction, not a separate per-render confirmation enforced by the current MCP tool.
- To stop using Hyference in that client, disconnect or remove the connector there; revoke its authorization too if your client offers that option. Removing a connector may only hide it locally. Hyference does not yet provide a connected-app revocation screen. Already accepted video jobs may still finish.
- Do not share download links publicly unless intended: anyone holding a valid link can access that video until the link expires.
Troubleshooting
API, keys & CLI — planned
There is no public REST API, personal API-key issuance or dedicated Hyference CLI yet. Do not use your app session as an API key. Coding assistants can use the prepared MCP connection if they support its OAuth flow; this is separate from a standalone command-line product.