Can ChatGPT or Claude generate video with MakeAIVideo?

Yes. The hosted MCP server at mcp.makeaivideo.ai connects to any MCP client over streamable HTTP with OAuth, so the assistant generates video as a tool call rather than by you pasting an API key into a script.

What does the agent get back, a clip or a finished video?

A finished video. One call returns an MP4 with the script written, the voiceover recorded, scenes generated, captions timed to the audio and music underneath. There is no editing step left for a human, which is what lets an agent workflow complete rather than stall.

How do I stop an agent spending all my credits?

Estimate first. estimate_video_cost is free and returns the exact credit cost before anything is created, and get_credits reports the live balance including in-flight reservations. API keys can also be scoped to read only, which removes the ability to spend entirely.

Is there a local option, or is it hosted only?

Both. The hosted server needs nothing installed. @makeaivideo/mcp on npm runs the same tools locally over stdio for clients that cannot reach an HTTP endpoint.

Does agent access cost extra?

No. MCP and API access are on every plan with no separate rate card, and renders draw on the same monthly credit allowance as the app. Plans start at $9 a month for 450 credits.

Which agent frameworks work with it?

Anything that speaks MCP, including Claude, Claude Code, ChatGPT in developer mode, Cursor and custom clients. Anything that cannot speak MCP can call the REST API directly with a bearer key.

Video generation for AI agents

Let an assistant produce a finished, publishable video rather than an asset that still needs editing. Connect over MCP in a minute, or call 56 REST endpoints directly.

Get an API key Connect over MCP

Why an agent finishes the job here

Most tools in this category return one component: a raw clip, a talking head, a voice track. An agent that receives a component cannot finish, because the remaining work is a human in a timeline. One call here returns the composed file, so the workflow completes.

Built for a caller that is not a person

Quote the cost before spending it

estimate_video_cost is free and takes the same arguments as the create tools, so an agent can show a price and get approval before committing. It turns an irreversible spend into a confirmable one, which is the property that makes autonomous use safe at all.

Orient itself for nothing

whoami, get_credits, list_voices, list_characters and list_templates all cost zero credits. An agent can check who it is, what it can afford and what is available without burning budget on exploration.

Retry without paying twice

POST /videos accepts an idempotency key. A retried request returns the original video instead of creating and billing a second one, so a flaky network does not become a double charge.

Poll without being throttled

Rate limits are tiered by cost rather than applied flat. Reads and status polls are unlimited; only generation itself is capped. An agent watching a render never trips a limit by waiting.

Fail in a way code can branch on

Every response carries a stable envelope with an error code and a request id, so an agent can distinguish a validation problem from an out-of-credits problem without parsing prose.

Get called instead of waiting

Webhooks push video.created, video.ready and video.failed to your endpoint, so a long render does not hold a conversation or a process open.

Start in one line

Point any MCP client at https://mcp.makeaivideo.ai and sign in with OAuth, or run @makeaivideo/mcp locally over stdio. Source for every package is on GitHub, MIT licensed.

Agent integration reference · MCP setup · TypeScript SDK · Developer hub

Agent FAQs

Can ChatGPT or Claude generate video with MakeAIVideo?: Yes. The hosted MCP server at mcp.makeaivideo.ai connects to any MCP client over streamable HTTP with OAuth, so the assistant generates video as a tool call rather than by you pasting an API key into a script.
What does the agent get back, a clip or a finished video?: A finished video. One call returns an MP4 with the script written, the voiceover recorded, scenes generated, captions timed to the audio and music underneath. There is no editing step left for a human, which is what lets an agent workflow complete rather than stall.
How do I stop an agent spending all my credits?: Estimate first. estimate_video_cost is free and returns the exact credit cost before anything is created, and get_credits reports the live balance including in-flight reservations. API keys can also be scoped to read only, which removes the ability to spend entirely.
Is there a local option, or is it hosted only?: Both. The hosted server needs nothing installed. @makeaivideo/mcp on npm runs the same tools locally over stdio for clients that cannot reach an HTTP endpoint.
Does agent access cost extra?: No. MCP and API access are on every plan with no separate rate card, and renders draw on the same monthly credit allowance as the app. Plans start at $9 a month for 450 credits.
Which agent frameworks work with it?: Anything that speaks MCP, including Claude, Claude Code, ChatGPT in developer mode, Cursor and custom clients. Anything that cannot speak MCP can call the REST API directly with a bearer key.