[ MCP ]Model Context Protocol

Let AI agents generate and publish your video

Connect ShortFast to Claude, ChatGPT, n8n or Make with one URL. The agent gets tools that mirror the REST API exactly, so it can write a brief, generate the video, render it and publish it to real accounts.

Read the MCP docsAPI and webhooks
[ 01 ]Connect

One URL, no package to install

The server is hosted, so there is nothing to run locally. Create an MCP key in the ShortFast app under Settings, then API & Webhooks, and paste the connection URL into your client.

Connection URL
https://app.shortfast.com/mcp?key=sf_mcp_xxxxxxxxxxxxxxxxxxxxxxxx

If your client sends headers instead of accepting a key in the query string, the same key works as a bearer token.

Header alternative
Authorization: Bearer sf_mcp_xxxxxxxxxxxxxxxxxxxxxxxx

Keys are scoped to one workspace, and a workspace holds at most one. An organization key can act on several workspaces by naming the one it should use:

Organization key, targeting one workspace
https://app.shortfast.com/mcp?key=sf_mcp_xxxxxxxxxxxxxxxxxxxxxxxx&workspaceId=<workspaceId>
[ 02 ]Clients

Works with the tools you already use

Any MCP-compatible client can connect. These four are the ones people ask about most.

Claude

Add it as a custom connector and the tools appear in the conversation.

ChatGPT

Paste the same URL where custom connectors are configured.

n8n

Drive generation from an existing workflow without writing HTTP nodes.

Make

Same connection URL, used as the MCP endpoint in a scenario.

[ 03 ]Capabilities

What an agent can actually do

Tools mirror the REST endpoints one to one, with identical validation and error codes.

01

Generate a video from a brief

Pick a pipeline, pass a prompt or a written script, and queue generation. The agent gets back a video with a queued status rather than a finished file.

02

Render and download

Queue the mp4 that turns a generated video into a file, poll its progress, and retrieve a download URL that is known to resolve.

03

Publish to connected accounts

Attach a rendered video to a publishing profile, choose platforms, and read back the publish state, including which platforms had no connected account.

04

Build and run automations

Create a blueprint from one brief that regenerates a fresh clip on a recurring schedule, check what is still blocking it, or run one clip immediately.

05

Edit the pieces a video is made of

Read and change the theme config, transcript, caption pages, b-roll segments, presentation timeline and AI scenes.

06

Check credits before spending them

Read what each feature charges and the ceilings the current plan enforces, so an agent can decide whether a job is affordable before queueing it.

Generation is asynchronous, and agents forget that

Queueing a video returns immediately with a queued status, not a finished file. The agent has to poll until the job reports COMPLETED or FAILED before it can render or publish anything.

This is the single most common thing to get wrong when wiring an assistant to a media pipeline. An agent that treats the queued response as success will confidently tell you a video is ready and hand you a URL that does not resolve yet. Worth stating explicitly in your system prompt.

Publishing spends real credits on real accounts

Generation and publishing draw on workspace credits and post to the social accounts you have connected. Those are not reversible in the way a read is, so they are worth a human confirmation step rather than leaving them to an autonomous loop.

Reading is safe: listing videos, checking an automation's health, or reading what a feature charges costs nothing and changes nothing. A sensible pattern is to let the agent plan and cost a job freely, then ask before it commits.

[ 04 ]FAQ

Common questions

What is the ShortFast MCP server?
An endpoint that exposes ShortFast over the Model Context Protocol, so an AI assistant can generate, render, publish and schedule video directly. The tools mirror the REST API one to one, with the same validation and the same error codes, so anything the API can do the agent can do.
Which clients can connect to it?
Claude, ChatGPT, n8n and Make connect by pasting the connection URL. Any MCP-compatible client works, since the server implements the protocol rather than a client-specific integration.
How does authentication work?
With an MCP key that starts with sf_mcp_, created in Settings then API & Webhooks in the ShortFast app. Pass it in the URL query string or as an Authorization: Bearer header. Keys are workspace-scoped and there is a maximum of one per workspace.
Can one key reach several workspaces?
An organization key can act on more than one workspace by appending a workspaceId parameter to the connection URL. A plain workspace key is limited to the workspace it belongs to.
Does the agent wait for the video to finish?
No. Generation and rendering are asynchronous, so the tool returns immediately with a queued status and the agent polls until the job reports COMPLETED or FAILED. Treat a queued response as work in progress rather than a finished video.
Is it safe to let an agent publish?
Generation and publishing spend workspace credits and post to real connected accounts, so both are actions worth confirming with a human before the agent runs them. Reading data is harmless; writing is not.
How is this different from the REST API?
It is the same surface reached a different way. The REST API is for code you write and deploy; MCP is for an assistant or an automation platform that discovers the tools at runtime. Teams commonly use MCP to prototype a flow and the API to ship it.
Does the MCP server cost extra?
No. It bills the same way the rest of ShortFast does: generation, rendering and publishing draw on workspace credits, and plan limits apply identically whether a call arrives over MCP or over HTTP.

Wire it up in a couple of minutes

Create an MCP key, paste one URL, and your assistant can generate and publish short-form video.

Start 3-day trial for $1See the full API