Shotstack can be driven directly from AI tools and coding agents, with no need to hand-write Edit JSON yourself. There are two ways to connect, depending on what you're using.
MCP Server: for chat clients and IDEs
A single endpoint at https://mcp.shotstack.io/ connects Shotstack to Claude, ChatGPT, Cursor, Claude Code, VS Code Copilot, Codex CLI, Gemini CLI, Windsurf, Zed, JetBrains, Goose, and Raycast. The agent composes the video edit, opens an inline preview (in chat clients) or returns a share link (in terminal clients), and renders on request.
Note: The MCP server is currently in beta, so tools and behaviour may change without notice.
To connect it, get an API key at app.shotstack.io, then add the connector in your client of choice (for example, in Claude Desktop: Settings → Connectors → Add custom connector → https://mcp.shotstack.io/).
Headless or CI use (no browser)
Most clients authenticate with OAuth, which needs a browser. If you're running an MCP-aware agent in a script, CI job, or anywhere else without a browser, you can skip OAuth entirely by passing your API key as an x-api-key header instead:
claude mcp add --transport http shotstack https://mcp.shotstack.io --header "x-api-key: YOUR_KEY"
Clients that only support stdio
The MCP server is now HTTP-only, so there's no local package to run directly. If your client can't speak HTTP and only supports local stdio servers, bridge to the remote endpoint with mcp-remote:
{
"mcpServers": {
"shotstack": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.shotstack.io/", "--header", "x-api-key:YOUR_KEY"]
}
}
}
For scripting and automation outside a chat client, the Shotstack CLI (see below) is usually a better fit than bridging the MCP server over stdio.
CLI + Claude Code Skill: for terminal agents
A terminal-native CLI plus a Claude Code Skill that gives coding agents (Claude Code, Cursor, Codex CLI, Gemini CLI, etc.) the same video-authoring conventions.
npm install -g @shotstack/cli
shotstack login # prompts for your API key, saves it for future sessions
shotstack --help
Add the accompanying skill with:
npx skills add shotstack/shotstack-cli
Common commands include shotstack render edit.json --watch to submit and poll a render, shotstack studio edit.json to open a draft for human review without spending render credits, and shotstack validate edit.json to check an edit against the schema before rendering.
Which one should I use?
If you're chatting with an AI assistant (Claude, ChatGPT) or working in an IDE, use the MCP Server. If you're scripting, running in CI, or working from a terminal-based coding agent, use the CLI, or the MCP server's header-based auth if you specifically need MCP tooling in a headless environment. Both follow the same underlying Edit JSON conventions, so the video output is consistent either way.