Guides

MCP Gateways for Image and Video Models: What fal and Sunra Offer, and What to Check Before Connecting an Assistant

20 October 2026 · RenderBob team

fal's hosted MCP server and Sunra's MCP page let an assistant run image and video models. The connection is also a new route for keys, prompts, and reference files.

Two services let an assistant run image and video models through MCP. That skips a round of dashboards. It also sends a key, a prompt, or a reference through another host.

fal

fal's blog of 19 March says the hosted server lets assistants search, run, and chain more than 1,000 models across image, video, audio, 3D, and upscaling. The post names Claude, Cursor, and Windsurf, and the setup steps include Claude Code and Claude Desktop. Nine tools cover search, a model's schema, pricing, docs search, run, submit, job status, file upload, and a recommendation. That post authenticates with a fal API key in the Authorization header. It says the key is never stored, each request is isolated, and the server is stateless. It does not mention rate limits, or a separate fee for the server. Model runs go through your fal key.

fal's current setup guide also offers an OAuth sign-in at mcp.fal.ai/mcp-relay, and it says that connection does not ask you to paste a key. The same guide adds job-result and cancel tools beyond the March nine, and it names ChatGPT and Codex among the clients. Check which URL you added before you assume the key never sits in a config file.

Sunra

Sunra's MCP page says an assistant can make image, video, audio, and 3D work after an OAuth sign-in, with no API key in the setup it describes. Credits are the same balance as the web app. The page says the assistant can check that balance and get an estimate, and that credits are spent only after you confirm, unless auto-confirm is on. Sunra says it does not use your prompts, uploads, or generated content to train its own models. Results land in a project on your Sunra account.

A daily digest on 20 October says Sunra charges each provider's API price with no markup. That sentence is not on Sunra's MCP page. Read the price in the account you are about to connect.

Why a studio would connect one

An assistant that can read a model's price and schema, submit a job, and check its status can compare models without opening six dashboards. That is the pattern we described for Runway on 15 October, and it sits next to the editor-side MCP notes in our Resolve and Palmier pieces.

Check these before you connect

• Where the file goes. An upload leaves your machine. Read retention and training terms for the gateway and for each model it calls. Sunra states a training rule for its own service. That sentence does not cover every model behind a gateway.
• Spending. Set a budget cap at the provider. On Sunra, leave auto-confirm off until you trust the loop.
• Key scope. If the connection wants a key, use a separate one and rotate it. fal's OAuth relay and a key in a header are different setups. Know which one you added.
• Terms per model. The gateway passes through each model's licence. Read the commercial terms of the models you actually run.
• Logs. Keep the prompts, references, and outputs for every clip you might deliver.
• A person looks at the result before anything is published or sent to a client.

When the references have to stay with the production, run that job on infrastructure the studio controls. RenderBob lets you choose local or cloud for each clip and keep the references with the studio.

What to watch

Whether either gateway publishes rate limits and data-handling terms in one place, and whether provenance survives a call that goes through the gateway.