Developers

MCP server Beta

Let an AI assistant work inside the project you have open — reading scenes, editing objects, and checking its work by sight, all over the Model Context Protocol.

Beta

MCP is in beta. It works today, but the tools and the way clients connect may still change as the standard matures.

MCP (Model Context Protocol) is an open standard that lets AI clients talk to apps on your own machine. SSHOW exposes the project you have open as a set of MCP tools, so an assistant like Claude can read its structure, make edits, and take screenshots to check the result — while you watch it happen on screen.

Note MCP runs only in the SSHOW desktop app. It opens a small server on your own computer (loopback only), which the browser editor can't do. This page applies to the desktop build, and MCP ships in both the Mac and Windows builds (Mac App Store and Microsoft Store editions included).

Get the desktop app

What the assistant can do

The assistant works through a small set of tools — it reads your project, makes edits, and checks its work by screenshot, and can also undo and redo, jump your window to a scene, and select what it changed. Every tool maps to something you could do by hand in the editor, nothing more.

Read your project Walks the scene and object tree, variables, and fonts to understand what's on the canvas before it touches anything.
Make edits Creates, changes, and arranges objects through the same commands you use — and each batch lands as a single Cmd+Z step, so anything is reversible.
Check by sight Renders a screenshot of any scene or object so the agent can check its work by sight — and a separate tool can switch the active scene to keep your screen in sync.

Turn it on

Four steps from a fresh install to your first AI edit.

  1. 1

    Open the desktop app

    MCP lives in the SSHOW desktop app. Install it first if you're working in a browser.

  2. 2

    Enable MCP

    Open Settings → MCP server and switch it On. Nothing listens until you do, and it stops the moment you switch it off.

  3. 3

    Connect your AI client

    Use Copy CLI command or Copy JSON config in Settings, then paste it into your assistant. SSHOW fills in the address and access token for you.

  4. 4

    Open a project

    The assistant only sees projects you have open. With several open, it works on the one you last focused — or you can ask it to switch to a specific project. Open the one you want and ask away.

Connecting a client

Two ways to register SSHOW, depending on your assistant. Both buttons appear in Settings → MCP server while it's running.

Claude Code — CLI command
Copy CLI command gives you a one-line claude mcp add command. Paste it into your terminal and Claude Code registers SSHOW as an HTTP server.
Claude Desktop & Cursor — JSON config
Copy JSON config gives you an mcpServers block that wraps SSHOW with the mcp-remote bridge — Claude Desktop reads only stdio servers from its config file, so the bridge turns the local HTTP server into one. Paste it into claude_desktop_config.json (Node.js / npx required), then fully quit and reopen the app — the config is read only at startup.

Safety & limits

MCP is built to stay private and reversible by default.

  • Off by default — the server starts only after you switch it on, and stops the moment you switch it off.
  • Local only — it binds to 127.0.0.1 (your own machine), so nothing on the network or internet can reach it.
  • Always reversible — every edit goes through the normal history, so you or the assistant can undo it.
  • No cloud — the bridge runs entirely on your computer; your project isn't sent anywhere to make this work.

Note Because the server listens locally, your AI client must run on the same machine as SSHOW. Remote or web-only clients can't connect.