Framesail works with Claude.
Connect it once and Claude makes the whole video — script, narrator voice, storyboard, every shot, final cut. You ask for a video; it comes back finished.
Connected · on autopilot
One prompt. One finished video.
No dashboard, no tabs. Claude calls the pipeline and the render comes back in the chat.
On it — making the video in Framesail.
Done — here's the finished video.
Connect it in a click.
MCP is an open protocol, so it isn't a Claude-only trick — any MCP client can drive the pipeline. Nothing to install.
claude.aiBrowser login
Claude DesktopBrowser login
Claude mobileBrowser login
Claude CodeOne command
- Cursormcp.json
- VS Code Copilotmcp.json
- Codex CLIconfig.toml
- Any MCP clientOpen protocol
claude mcp add --transport http framesail \
https://api.framesail.com/mcp \
--header "Authorization: Bearer fsk_..."OAuth clients skip the key entirely — point them at api.framesail.com/mcp and sign in. Full setup is in the docs.
Connect it once,
then just ask for a video.
One prompt. Or every gate.
Most people just say “build it” and read the result. If you'd rather sign off on each stage first, ask for that instead — Claude waits.
Say it once
“Build me a channel about deep-sea mysteries and make the first video.”
- script
- references
- voiceover
- storyboard
- render
Claude calls every stage in order and comes back with the finished video.
Or hold the gates
“Same — but show me the script and the storyboard before you render anything.”
- scriptapproved
- referencesapproved
- voiceoverapproved
- storyboardyour call
- render
It stops where you told it to and waits. Nothing renders until you say go.
What Claude can do once it's connected.
Give it a niche, let it research the space, then send it off to build the channel. All of this runs on its own — and every one of them is also a thing you can ask for by name.
SCENE 03 — THE LAUNCH PAD
Then the countdown began.
Nine minutes out, the pad crew cleared the tower.
SCENE 04 — MISSION CONTROL
Write and revise the script
Ask for a topic — or rewrite one scene without touching the rest.


Lock the cast and the world
Characters and places fixed once, before a single shot renders.
- Video
Veo 3.1
ImageNano Banana Pro
VoiceElevenLabs v3
Pick the model per job
Veo, Kling, Seedance, Nano Banana, ElevenLabs — set per job or let the default ride.

Re-render
Direct the shots
Reorder the board, re-render one bad frame, turn a still into a clip.
- scriptreferencesvoiceoverstoryboardeditablesegments
Read every stage
Script, references, storyboard, renders — all readable between calls.
https://framesail.com/v/…final.mp4Export the finished video
One call renders the cut and hands back a URL.
One brief in, a finished video out.
Claude calls the steps in order. State lives server-side, so a run can pause for hours and pick up in a fresh chat — and you can step in at any stage without restarting it.
Prefer code? There's a REST API.
The MCP tools are thin wrappers over REST endpoints — same pipeline, same authentication. Script it yourself or wire it into your own product.
API referenceQuestions about claude & MCP.
How do I use Framesail in Claude?
Add Framesail as a custom connector in claude.ai — the URL is https://api.framesail.com/mcp — then log in and approve. Framesail appears in Claude's tools, and from there you just ask: "make a video about X." Claude calls the pipeline for you: script, references, voiceover, storyboard, render. The same connector works in Claude Desktop and the mobile app. For Claude Code or Cursor, add the server with an fsk_ API key instead.
Is the MCP server remote or local?
Remote. It's a streamable-HTTP MCP server at the Framesail API — nothing to install or run on your machine. OAuth-capable clients like claude.ai and Claude Desktop connect with a browser login; Claude Code and Cursor connect with an fsk_ API key.
What can Claude actually make?
Finished long-form video, not clips. Claude drives the whole pipeline — script, characters and settings, narrator voiceover, storyboard, animated segments, final render — and hands back a video URL. It can also go back and change one scene, re-render a single shot, or swap the voice without restarting the run.
Does it work with agents other than Claude?
Yes. MCP is an open protocol, so any MCP client works — Cursor, Claude Code, and others. And there's a plain REST API underneath if you'd rather script it yourself or wire it into your own product: same pipeline, same authentication, no MCP client required.
How does authentication work?
Two ways. OAuth login: claude.ai, Claude Desktop, and other OAuth-capable MCP clients connect to the URL and pop a browser login — no key handling. API keys: create a long-lived fsk_ key in the web app and send it as an Authorization: Bearer token for Claude Code, Cursor, or REST. Both show up as revocable keys on your account page, and agents can't mint or revoke keys — only a signed-in browser session can.
Is there an AI video generator API for long-form video?
Yes — the MCP server and the REST API are two surfaces over the same pipeline. Most video generation APIs return single clips; Framesail's runs the whole pipeline — script, locked references, storyboard, voiceover, final render — and returns a finished long-form video. One call chain instead of stitching clip APIs together yourself.
Put Framesail in your Claude.
Connect it once, then ask for a video the way you'd ask for anything else.