AI agents (MCP)
Overview
Popcorn Editor ships a built-in MCP server — one URL any MCP-capable AI agent can connect to and design on your behalf. Claude, ChatGPT, Gemini CLI, Cursor, or your own agent connects over streamable HTTP, authenticates as you, and gets a small set of tools for creating, reading, and updating designs in your workspace:
https://app.example.com/mcp/popcorn
Your exact URL is shown in the app under Settings → Your agent access.
MCP usage consumes no AI credits. The connected agent brings its own model — Popcorn only validates and compiles documents server-side.
What agents can do
Agents work in the Design Document format (DDS) — never raw editor state. Everything is scoped to your workspace and permission-checked server-side: agents can't read other users' data or touch admin functions.
- Create designs — validate and compile a Design Document into a new design, returning an editor URL you can open immediately.
- Read and update existing designs — designs are read back as Design Documents; updates snapshot the previous version first, so every agent edit is restorable from version history.
- Self-check before writing — a free, side-effect-free validation tool lets agents loop until their document is correct.
- Start from templates — browse the template gallery and spin up a design from any accessible template.
- Learn the format on the fly — the server publishes the Design Document JSON Schema, an authoring guide, and a guided authoring prompt as MCP resources.
Full details for every tool, resource, and prompt: Tools reference.
Two ways to connect
| OAuth sign-in | Agent token | |
|---|---|---|
| Best for | App connectors: Claude.ai web, Claude Desktop, ChatGPT | CLIs and IDEs: Claude Code, Gemini CLI, Cursor, VS Code |
| Setup | Paste the URL, sign in, click Authorize | Mint a token in Settings → Your agent access, send it as an Authorization: Bearer header |
| Access | Read and write | Read & write or read only (your choice when minting) |
| Lifetime | 24-hour tokens, refreshed automatically for up to 30 days | Optional expiry: 30, 90, or 365 days |
| Revoke | Remove the connector in the AI app | Delete the token in Your agent access |
For token scopes, expiry, and security details, see Tokens & security.
Pick your client
- Connect Claude — Claude.ai web, Claude Desktop, and Claude Code
- Connect ChatGPT
- Connect Gemini CLI
- Other clients — Cursor, VS Code, and any other MCP client
New to the idea? Start with the walkthrough: Let AI agents design for you.
Limits and admin control
Requests are rate-limited to 60 per minute per user. Administrators can disable the entire MCP server instantly under Admin → Settings → External agent access (MCP) — useful as an organization-wide kill switch.