MotionVid for AI agents

Endpoint

The server speaks Streamable HTTP at:

https://motionvid.ai/mcp

Authentication is OAuth 2.1 with automatic discovery: point a client at the URL, click Connect, and sign in with your MotionVid account. Headless setups can send an API key instead (below).

Connect a client

ChatGPT

Click Install plugin, sign in with your MotionVid account, then type @Motionvid in any chat.

Claude (web & desktop)

1. Customize → Connectors → + Add → Add custom connector
   (claude.ai/customize/connectors, same in the desktop app)
2. Name: MotionVid — URL: https://motionvid.ai/mcp
3. Click Connect and sign in with your MotionVid account

Claude Code

# 1. Register the server:
claude mcp add --transport http motionvid https://motionvid.ai/mcp

# 2. Sign in (adding it does not start OAuth on its own):
claude mcp login motionvid
#    …or run /mcp inside a session and choose Authenticate.

# Headless alternative — register with an API key, no sign-in step:
claude mcp add --transport http motionvid https://motionvid.ai/mcp \
  --header "Authorization: Bearer mvk_your_api_key"

Cursor

{
  "mcpServers": {
    "motionvid": {
      "url": "https://motionvid.ai/mcp",
      "headers": { "Authorization": "Bearer mvk_your_api_key" }
    }
  }
}

VS Code

{
  "servers": {
    "motionvid": { "type": "http", "url": "https://motionvid.ai/mcp" }
  }
}

Codex

[mcp_servers.motionvid]
url = "https://motionvid.ai/mcp"

# Headless alternative — this names the variable, so also
# export MOTIONVID_API_KEY=mvk_your_api_key in your shell:
# bearer_token_env_var = "MOTIONVID_API_KEY"

# Sign in (adding the server does not start OAuth on its own):
codex mcp login motionvid

# Or skip the config file and do both steps from the terminal:
codex mcp add motionvid --url https://motionvid.ai/mcp

Any MCP client

# Streamable HTTP + a bearer header is all it takes.
# OAuth access tokens and mvk_ API keys are interchangeable.

curl -X POST https://motionvid.ai/mcp \
  -H "Authorization: Bearer mvk_your_api_key" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Example tool call

Your agent handles this for you — but under the hood, generating a video is one call:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "create_video",
    "arguments": {
      "prompt": "A 30s product launch teaser for our new app",
      "duration": 30,
      "aspect_ratio": "16:9",
      "render": true
    }
  }
}

Leave a setting out and the build decides it from your prompt and whatever you attach — “for Reels” goes portrait, a script sets its own length — landing on 30 seconds, 16:9, narrated only when nothing says otherwise. Anything you pass explicitly is binding. Your agent is told to ask you first when the request is thin on length, shape, or narration and the answer would change the result, so a vague prompt gets a quick question rather than a guess.

Drop a link in the prompt and the video follows its design. Any URL or domain you mention gets read and screenshotted, and the page's copy, colors, and typography carry into the scenes — so pointing at your own site or the article you're announcing is the quickest way to make the result look like it belongs to you.

To match the look of a video you admire, pass it in reference_videos — an Instagram, TikTok, YouTube or X link, or a link to the file. The build studies its palette, type, layout and pacing before it starts, then makes your video in that style, never copying its subject or showing the clip itself.

It also researches on its own when the prompt needs it. Ask for something current and it searches the web before planning, so a video about this week's news is built on this week's sources rather than guesswork. Name a company without a link and it looks up the official site itself.

Available tools

Availability & billing

MCP is available on every paid plan. API keys for headless use are separate, and come with Ultimate and Creator; create them from your API keys page. Generations started by an agent spend the same credits as the web app, and research along the way costs one credit for each page read and each web search run.

Connecting a client grants it access to your MotionVid account — your videos, brand kits, and credit balance. Every tool is marked read-only or account-changing, and MCP clients that respect that labeling, including Claude, ask for confirmation before an account-changing tool runs. See AI Agent Connectors and MCP in our Privacy Policy for what a connected client can access and how to revoke it, and the Terms for how agent-initiated use is treated.

Support

Questions about the MCP server, a connector that isn't behaving, or anything else — email nick@motionvid.ai.