Setup

Fastest setup — one line

Paste this into your agent's terminal (Claude Code, Cursor, Codex) and it wires itself up:

npx -y cueframe install

It detects every agent on the machine — Claude Code, Claude Desktop, Cursor, Codex, VS Code — and wires the CueFrame MCP server into each, then installs the agent skills. No API key. On the first call a browser opens and you click Allow; sign-up happens there if you're new. That's the whole setup.

Or, from an agent that can browse, paste set up https://cueframe.ai/skill.md and it runs the same install for you.

Prefer to wire it by hand? CueFrame is one service behind three front doors. Pick the one that matches how your agent runs, sign in once, and every tool, command, and endpoint resolves to the same account.

Authentication

The default is browser sign-in — no API key. MCP clients and the CLI log you in through the browser: add the server (or run cueframe login), a browser opens, you click Allow, done. Nothing to paste, nothing to store. Best for anything interactive.

Advanced — API keys (CI / headless / REST). For non-interactive clients that can't open a browser, mint a cf_live_… token in the console and send it as a bearer header. Scope it (read-only vs. write) and revoke any that leak.

Authorization: Bearer cf_live_xxxxxxxxxxxx

The v1 API base is https://api.cueframe.ai/v1.

Or skip the account entirely. An autonomous agent can pay per request with a wallet — no key, no signup. See Pay per call.

MCP

The hosted MCP endpoint — https://api.cueframe.ai/v1/mcp — lets any MCP client drive CueFrame's curated tool surface directly.

Browser sign-in (no API key)

Add the server URL with no key. On first use your client opens a browser — sign in to CueFrame and click Allow. That's the whole setup:

claude mcp add --transport http cueframe https://api.cueframe.ai/v1/mcp

Or as JSON config:

{
  "mcpServers": {
    "cueframe": { "type": "http", "url": "https://api.cueframe.ai/v1/mcp" }
  }
}

With an API key (headless / CI)

For non-interactive clients that can't open a browser, pass a key as a header instead of signing in:

claude mcp add --transport http cueframe https://api.cueframe.ai/v1/mcp \
  --header "Authorization: Bearer cf_live_YOUR_KEY"

Mint the key in the console or with cueframe keys create (below).

Once connected, head to the Quickstart to go from intent to a rendered video in three steps.

CLI

The cueframe CLI turns media into rendered video from the terminal. Every command speaks --json (an NDJSON event stream) and self-describes via cueframe describe --json, so agents can drive it without guesswork.

Install

npx cueframe@latest --help      # zero-install
# or
npm i -g cueframe               # global `cueframe`

Sign in

cueframe login                       # browser sign-in — no key
# or
cueframe auth cf_live_YOUR_KEY       # static API key (CI / headless)

The key can also be supplied via the CUEFRAME_API_KEY environment variable. After cueframe login, mint keys for other surfaces without leaving the terminal:

cueframe keys create --name ci

Then follow the Quickstart for the upload → author → render loop.

REST

Every surface resolves to the same v1 REST API. REST is key-only — call it directly with a Bearer token:

curl -X POST https://api.cueframe.ai/v1/renders \
  -H "Authorization: Bearer $CUEFRAME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "projectId": "..." }'

See the API reference for every endpoint, including POST /v1/api-keys to mint keys programmatically.