Relo speaks MCP, so Claude, Cursor and every other AI client can drive it directly. Describe the video in chat, watch the preview come back in the conversation, change it by saying what is wrong, then render and download — on the same credits, brand kit and voice as the website.
Works with Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, Cline, Gemini CLI and Codex CLI.
On this page
You talk to your AI client the way you already do. Nothing here is a command to memorise — the tools describe themselves, so the assistant knows when to reach for them.
Two things are worth noticing. The assistant quotes the price and waits for a yes before spending anything. And you never asked to see the video — the frames arrive on their own, the moment it is ready, the same way the preview appears on the website.
Two minutes, once. After that every project on this machine can make videos.
Keys live on your API keys page. Give it a name you will recognise later — the name is the only thing that tells two keys apart when you come back to revoke one. The key is shown once and stored hashed, so copy it when it appears.
For Claude Code, that is one command:
claude mcp add relo \ --env RELO_API_KEY=relo_sk_YOUR_KEY \ -- npx -y @relo-video/mcp
Everywhere else it is the same block of JSON. Only the file it belongs in changes:
{
"mcpServers": {
"relo": {
"command": "npx",
"args": ["-y", "@relo-video/mcp"],
"env": { "RELO_API_KEY": "relo_sk_YOUR_KEY" }
}
}
}VS Code names the object servers rather than mcpServers and wants an explicit transport, in .vscode/mcp.json:
{
"servers": {
"relo": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@relo-video/mcp"],
"env": { "RELO_API_KEY": "relo_sk_YOUR_KEY" }
}
}
}Codex CLI uses TOML, in ~/.codex/config.toml:
[mcp_servers.relo]
command = "npx"
args = ["-y", "@relo-video/mcp"]
env = { RELO_API_KEY = "relo_sk_YOUR_KEY" }Any other MCP client works too. However it asks for its servers, give it the command npx, the arguments -y @relo-video/mcp, your key in RELO_API_KEY, and stdio as the transport.
You should see ten relo_* tools. Then it is just English: “make me a 20-second promo for my app, vertical, with captions.”
You will rarely name these yourself — the assistant picks them. They are listed so you know exactly what it can and cannot do on your account.
MCP has no video content type — no AI client can play a clip inside the conversation. Rather than pretend otherwise, the preview arrives in three layers, and you will usually only need the first.
The moment the video is ready, still frames sampled across it appear in the chat. You did not ask; they arrive. Enough to judge the headline, the type size, the colours and the structure.
Ask to play it and your default video player opens with the real thing — motion, timing and sound. This is the one to use when you care about pacing.
Previews carry a relo.video mark in the corner. It is a draft marker: the final render is clean, and the mark is stripped automatically when you render. There is nothing to remove and no setting to change.
Every preview comes with a link to the video’s own page on Relo. It sits behind your login, so it opens for you and nobody else.
A video is only half the job — it still has to go out. PostCraft is our open-source posting tool, MIT-licensed and free, and it is the fastest way to get more out of Relo MCP than one clip at a time. It runs entirely on your own machine: it drafts posts in your voice for LinkedIn, X, Instagram and TikTok, calls Relo over MCP whenever a post needs a video, and schedules the lot through browser sessions you have already logged into.
The point is the loop. Relo makes the video; PostCraft decides what the week should say, which posts need motion, and when each one goes live — so a single brief becomes a fortnight of scheduled content instead of a file on your desktop.
A LinkedIn post and an X post are different shapes. PostCraft drafts them separately, and picks the medium each one wants — plain text, an image, a carousel, or a Relo video.
Rewrite a draft and the reason is recorded in content/preferences.md. The next batch already knows.
Posts go out through your own logged-in browser — no passwords stored anywhere — and every step is screenshotted, so a failure leaves you a picture of what it saw.
Anything it cannot know is left as a [[placeholder]] for you to fill, rather than a plausible-looking figure.
node calendar.cjs opens a browsable calendar of what is queued, in your audience’s timezone.
Clone it, install, and open the folder in Claude Code. It needs Node and Chrome, and nothing else.
git clone https://github.com/Relo-video/PostCraft postcraft cd postcraft npm install
Then three commands do everything: /setup-social once, to capture your voice, your platforms and your posting rhythm and to log you in; /create-posts to draft a batch; and /schedule-posts to put them in the queue.
Add Relo to the same project and PostCraft will use it for any post that wants a video:
claude mcp add relo \ --env RELO_API_KEY=relo_sk_YOUR_KEY \ -- npx -y @relo-video/mcp
PostCraft works without Relo connected — it simply writes text-and-image posts instead. One thing to know before you start: scheduling through a browser is against most platforms’ terms of service, and the repo says so plainly. Read the README and decide for yourself.
Just ask for the one you want — “make it a hand-drawn story”, “turn this recording into a walkthrough”. Claude picks the mode; none of them needs files unless it says so.
Describe it and Relo writes the script, designs the scenes and animates them. Your clips, screenshots and audio are optional.
Attach a screen capture: it is framed on a designed stage, zoomed into each action, labelled and broken into steps.
Hand-drawn animation — drawn characters, people, animals or mascots, act out your product’s story in drawn places.
The same hand-drawn story staged in 3D space, with layered scenes and a camera that moves through them.
Bring your own character. In the 2D and 3D story modes, attach your mascot or character art and it stars in the film, animated as it is — PNG, JPG, WebP, SVG, or the design file itself (.ai, .eps, .pdf, .psd). A photo of a person or pet is redrawn in the film’s style. Attach nothing and Relo designs the cast.
Exactly what it costs on the website. The credits come out of the same balance and appear in the same history — a video made from Claude is not billed differently from one made in the browser.
| What | Detail | Credits |
|---|---|---|
| A video | Script, cuts, motion graphics, the lot | 35 |
| A 2D or 3D story | Hand-drawn animation; voiceover always included (+10) | 50 |
| An edit | Any plain-language change, as many as you like | 5 |
| A render | Per aspect ratio — two shapes costs 20 | 10 |
| AI voiceover | One of 21 built-in narrators | +10 |
| Your own cloned voice | Replaces the voiceover charge, not added to it | +20 |
| Background removal | Cut the subject out of your footage | +10 |
| Captions, music, stock, brand kit | Included | free |
Both are one-time uploads, and neither can be done from a chat window — they need a file from you.
Your cloned voice. Record eight to fifteen seconds of yourself in Options → My voice. After that any video can be narrated in your voice instead of a stock one.
Your brand kit. Fill in your product name, description, colours and logo on your profile. Videos then come out in your brand rather than a generic one.
Ask for either before it exists and the tool stops and tells you how to set it up. It will not quietly make the video without the thing you asked for.
Handing an AI assistant a key that can spend money deserves a straight answer about the guardrails. All of these are enforced on Relo’s servers, not in the package on your machine, so they hold however the tools are called.
Every cost is calculated server-side by the same code that charges browser users. The package cannot compute or discount a price of its own.
Ask for a cloned voice you have not recorded, screen-recording mode with no recording attached, or a design file outside the 2D / 3D story modes, and it stops and explains — with nothing deducted.
Minting and revoking stay behind your browser login, so one leaked key cannot grow itself a replacement.
Delete a key on your profile and the next call from it fails.
If generation breaks, the credits come back automatically and the status says so.
Restart the client — MCP servers are only read at startup. If they are still missing, check the config file is valid JSON and that the key has no stray quotes around it.
The key is wrong, or it has been revoked. Create a fresh one on your API keys page and update the config.
A minute or two is normal; ten is not. Ask for the status — if it has failed, the credits have already been returned.
Client makers move these around. Check that client’s own MCP documentation for where its server config lives — the command, arguments and environment stay the same wherever it goes.
Your next video, from the terminal you are already in.
New accounts start with 100 free credits — enough for two videos and change.
Create your API key →