MCP server

Use AIVeed from Claude, Cursor, VS Code or any other Model Context Protocol client. The open-source aiveed-mcp server gives your assistant tools to check prices, generate videos, images and audio, wait for the result and hand you the hosted URL — paid in credits from your own account.

It talks only to the public API at /v1 with your API key, so it can do nothing your key could not already do with curl. Source, security model and releases: github.com/AIVeed-io/aiveed-mcp (MIT). Every npm release is built by GitHub Actions and published with provenance.

1. Get an API key

Create one on the developer dashboard. Node.js 20 or newer must be installed; npx downloads the server on first use.

2. Add the server to your client

Claude Code

bash

claude mcp add aiveed -e AIVEED_API_KEY=aiv_sk_... -- npx -y aiveed-mcp

Claude Desktop and Cursor

Add this to claude_desktop_config.json (Claude Desktop: Settings → Developer → Edit Config) or ~/.cursor/mcp.json (Cursor), then restart the app.

json

{
  "mcpServers": {
    "aiveed": {
      "command": "npx",
      "args": ["-y", "aiveed-mcp"],
      "env": { "AIVEED_API_KEY": "aiv_sk_..." }
    }
  }
}

VS Code

Add this to .vscode/mcp.json in your workspace, or run MCP: Add Server from the command palette.

json

{
  "servers": {
    "aiveed": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "aiveed-mcp"],
      "env": { "AIVEED_API_KEY": "aiv_sk_..." }
    }
  }
}

Any other MCP client works the same way: run npx -y aiveed-mcp over stdio with AIVEED_API_KEY in its environment. Pin a version (aiveed-mcp@0.1.0) if you want upgrades to be deliberate.

Keep your client's approval prompt on for the generate_* and extend_video tools: they spend credits. Ask the assistant to check the price with list_models first.

3. Ask for a video

prompt

Using AIVeed, make an 8-second 16:9 video of a paper boat drifting down a rainy street at night.
Tell me the cheapest model that fits and its credit cost before you generate, then wait for it and give me the link.

Tools

Tool	What it does	Credits
`list_models`	Live model catalog: options, accepted inputs and prices (no key needed)	free
`get_credits`	Your credit balance	free
`generate_video`	Start a video job; returns a job id	spends credits
`generate_image`	Start an image job	spends credits
`generate_audio`	Start a voiceover or music job	spends credits
`extend_video`	Add ~7 s to a completed Veo 3.1 video	spends credits
`get_job`	Job status; `wait_seconds` polls until it finishes	free
`list_jobs`	Your jobs, filterable by status, model and type	free
`list_videos / list_images`	Your generated media, with permanent URLs	free
`cancel_job`	Cancel a queued job (refunded)	free
`upload_media`	Turn a local image, audio or video file into an input URL (local setup only, off by default)	free

Every generate call sends an Idempotency-Key and returns it, so a retry with the same key never charges twice. Failed jobs are refunded automatically.

Local files

upload_media turns a file on your computer into a URL a model can use (a first frame, a reference image, audio for lipsync). The URL is public, so the tool is off until you list the folders it may read in AIVEED_MCP_UPLOAD_DIRS (comma-separated; ~ is your home folder):

json

"env": {
  "AIVEED_API_KEY": "aiv_sk_...",
  "AIVEED_MCP_UPLOAD_DIRS": "~/Videos/aiveed-inputs"
}

List a folder that only holds media you are happy to share. Even inside it, the server never reads hidden files or folders, system directories or linked files, and only uploads real image, audio and video files, checked by their content. See Uploads for the size limits.

Hosting it over HTTP

To run one server for several people, start it in Streamable HTTP mode behind a TLS proxy:

bash

npx -y aiveed-mcp --http --port 8787 --host 127.0.0.1
# clients then call POST http://127.0.0.1:8787/mcp with  Authorization: Bearer aiv_sk_...

Each request must carry its caller's own key, so a hosted server never spends anyone else's credits, and upload_media is not offered. The README lists the host, origin and limit settings.

Prefer plain HTTP?

Agents that can make HTTP requests can use the API directly: see the prompt packs and llms-full.txt.