docs /mcp

Let an agent work inside Shyne.

Connect an AI client to your projects over MCP — the hosted web endpoint, or the desktop app’s local endpoint. Includes a block you can hand straight to your agent.

MCP is the direct route from a compatible AI client to Shyne. It exposes project discovery, artifact editing, visual verification, media, video, and safe scheduling tools. Connect to the hosted web endpoint, or to the desktop app’s local endpoint on your own machine.

What MCP does

The hosted endpoint authenticates with OAuth and applies project permissions. The desktop endpoint uses the signed-in app session and has the local-machine trust boundary described below.

Connect

Shyne has two MCP endpoints. Most clients use the hosted web endpoint; the desktop endpoint connects an agent to the app running on your own computer.

The hosted endpoint works from any client, anywhere, and connects to your Shyne cloud projects.

MCP endpoint
https://mcp.shyne.ai/mcp
Claude Code
claude mcp add --transport http shyne https://mcp.shyne.ai/mcp --scope user

Add this URL as a remote HTTP MCP server in your client. The first connection opens a Shyne sign-in flow. OAuth credentials stay in the MCP client.

Hand this to your agent

Paste this into any AI client — it connects, orients, and starts working. A person can read it too.

paste this into your agent
You are working with Shyne (shyne.ai), a collaborative design workspace.

1. Connect to the Shyne MCP server: https://mcp.shyne.ai/mcp (Streamable HTTP, OAuth).
2. Call projects.list and match the user’s project by exact name or id. Ask if unsure.
3. Call get_guide once. It returns the current authoring rules; follow them over anything cached.
4. Use tools/list as the live source of tool names and schemas.
5. Work one frame at a time. Read a frame immediately before editing it, and screenshot every frame you change.

Full guide: https://shyne.ai/docs/agent.md · Discovery: https://shyne.ai/llms.txt

The live authoring rules come from the get_guide tool after connecting; they are versioned with the product, so never rely on a cached copy. Machine-readable references (/llms.txt, /docs/agent.md, openapi.json) are listed on the Reference page.

The working loop

OrientReadChangeLookReport

A connected agent first orients: it lists your projects, loads the current working guide with get_guide, and looks at what is already on the page. Then it edits one frame at a time and screenshots each change to check its own work before reporting back.

Permissions

An authenticated agent can work only in projects owned by that Shyne user. Destructive operations are separate from normal writes. Whole-project deletion, credential management, final publishing, and provider tokens are not exposed to the agent. The desktop endpoint adds no separate sign-in: it is loopback-only and acts as the user signed into the desktop app.