1. ultrafast
  2. Integrations

Integrations

Wherever your agent runs.

Coding agents, any MCP client, Python and TypeScript, an HTTP API and Docker, all on your own GPU today. The ChatGPT app and the Claude connector come with ultrafast Cloud.

Coding agents

Claude Code, Codex, Cursor, Gemini CLI.

Run uf mcp on your machine and the agent gets 20 tools: it writes the document, looks at frames and contact sheets, fixes what review finds and exports.

Claude Code

Claude Code
claude mcp add ultrafast -- uf mcp

Codex

~/.codex/config.toml
[mcp_servers.ultrafast]
command = "uf"
args = ["mcp"]

Cursor

.cursor/mcp.json
{
  "mcpServers": {
    "ultrafast": { "command": "uf", "args": ["mcp"] }
  }
}

Gemini CLI

~/.gemini/settings.json
{
  "mcpServers": {
    "ultrafast": { "command": "uf", "args": ["mcp"] }
  }
}

Each agent also gets the ultrafast skill: npx skills add ultrafastmov/ultrafast. In Claude Code, the plugin brings both: claude plugin marketplace add ultrafastmov/ultrafast, then claude plugin install ultrafast@ultrafast.

Remote MCP

From your own server.

uf serve answers MCP over Streamable HTTP at /mcp, with its API key as a bearer token. The tools take documents and asset ids instead of file paths; frames and contact sheets come back as images, so the agent sees its work.

A hosted server with OAuth sign-in comes with ultrafast Cloud.

MCP docs

Claude Code
claude mcp add --transport http ultrafast \
  http://localhost:8080/mcp \
  --header "Authorization: Bearer $ULTRAFAST_API_KEY"
any MCP client
{
  "mcpServers": {
    "ultrafast": {
      "url": "http://localhost:8080/mcp",
      "headers": { "Authorization": "Bearer uf_live_..." }
    }
  }
}

SDKs

Python and TypeScript.

render, frame, sheet, check, review, estimate and upload, against your own uf serve. Local footage, images and fonts in the document are uploaded for you.

pip install ultrafastmov
from ultrafastmov import Ultrafast

# your uf serve, and the key it printed
uf = Ultrafast(base_url="http://localhost:8080")

doc = uf.load("film.json")
for issue in uf.review(doc):
    print(issue.kind, issue.node, issue.message)
uf.render(doc).save("film.mp4")
npm i ultrafastmov
import { Ultrafast } from "ultrafastmov";

// your uf serve, and the key it printed
const uf = new Ultrafast({ baseUrl: "http://localhost:8080" });

const doc = await uf.load("film.json");
await (await uf.sheet(doc, { count: 9 })).save("sheet.png");
const render = await uf.render(doc);
await render.save("film.mp4");

Agent frameworks

Starters for the frameworks you build on.

Starter apps for the Vercel AI SDK, the OpenAI Agents SDK, LangGraph and Mastra: an agent that takes a brief, works through the /mcp tools of your uf serve and returns a reviewed MP4. They're in the starters/ folder of the repo.

Starters on GitHub

OpenAI Agents SDK
from agents import Agent, Runner
from agents.mcp import MCPServerStreamableHttp

async with MCPServerStreamableHttp(params={
    "url": "http://localhost:8080/mcp",
    "headers": {"Authorization": f"Bearer {key}"},
}) as ultrafast:
    agent = Agent(
        name="Editor",
        mcp_servers=[ultrafast],
        instructions="Make it, review it, export it.",
    )
    result = await Runner.run(agent, "A 15 s launch promo")

HTTP API

One request, one video.

POST a document to your uf serve, get an MP4 back.

HTTP API docs

curl
jq '{document: ., output: {inline: true}}' film.json |
curl http://localhost:8080/v1/render \
  -H "Authorization: Bearer $ULTRAFAST_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- -o film.mp4

Self-host

Your GPU.

The render service is open source: the HTTP API and remote MCP on one machine with a GPU, one API key, files on local disk. Enough to build against and to share one box on a small team.

Jobs with signed webhooks, output to your own bucket, sign-in and GPUs that scale are planned for ultrafast Cloud.

SELF_HOST.md

Docker
docker run --gpus all -p 8080:8080 -v ultrafast:/data \
  ghcr.io/ultrafastmov/ultrafast:latest serve

CLI and live player

And a shell.

uf live film.json plays a document while you or an agent edit it; parameter edits show on the next frame.

CLI reference

shell
uf new vertical film.json
uf check film.json
uf live film.json
uf review film.json --safe social
uf export film.json film.mp4

Chat apps · coming soon

ChatGPT and Claude.

For anyone who can describe a video. They render on ultrafast Cloud, so they open with it.

Join the waitlist

ChatGPT app Coming soon

  • Add ultrafast from ChatGPT's app directory.
  • Ask for a video; attach footage, photos or a logo.
  • Edit it by asking, then download the file.

Claude connector Coming soon

  • Add ultrafast in Claude's settings under Connectors, then sign in.
  • Works in Claude on the web, desktop and phone.
  • The ultrafast skill, which teaches the document and the review loop, works today in Claude Code.

Pick one and start.

Every integration renders the same documents with the same engine.

Read the quickstart

Want it hosted? Join the ultrafast Cloud waitlist.