FISYCO runs an MCP server that gives AI coding agents your design system. Before the agent writes UI, it can find the components you already have, read how your team calls them, look up token names and check its code against the same notes your pull request check leaves. This page shows how to connect Claude Code, Cursor, VS Code with GitHub Copilot or Windsurf.
Before you start
- You have a FISYCO account with access to the project the agent will work in.
- The design system in code is set up for the project. See Connect the design system in code. Without an indexed repository the agent still gets your tokens and team rules, but not your components.
Connect your agent
The server is remote, so there is nothing to install. Your agent connects to one address and signs in through your browser.
- Open the account menu, then Account settings → AI agents.
- Under 1. Add FISYCO to your agent, pick your agent's tab and copy the snippet. It already holds the right address.
- Run it in a terminal or add it to the file the tab names, then start your agent.
- The first time it connects, the agent opens a FISYCO page in your browser. Check the agent's name and where the answer goes, then click Allow.
The snippets on the page look like this:
GitHub Copilot's agent mode in VS Code reads .vscode/mcp.json, so use the VS Code tab for Copilot.
The consent page warns you for a reason. Click Allow only if you started connecting this agent yourself, just now. If someone sent you the link, click Deny.
The agent then acts as you. It sees only the workspaces and projects you can see, with your role.
Sign in with a token instead
An agent that can't open a browser, such as one in CI or on a remote machine, can send a personal access token. Create one on the Access Tokens page; after you create it, FISYCO shows a config with the token in the Authorization header:
{
"mcpServers": {
"fisyco": {
"type": "http",
"url": "https://api.fisyco.com/mcp",
"headers": { "Authorization": "Bearer <your token>" }
}
}
}The tools that read your design system in code need the codebase read scope, and the Access Tokens page doesn't offer it. With the matching scopes ticked, a token made there covers the Figma library tools, tokens and the design system guide. For the other tools, including component search and check_ui_code, use the browser sign-in.
More on scopes and expiry in Access tokens.
What the agent gets
Fourteen tools. All of them only read, except the one that starts a library sync. Each tool's description tells the agent when to call it.
check_ui_code reads the fragment and doesn't store it. The rules your team sets for agents reach them through get_design_system_rules, search and check_ui_code; see Rules for AI agents.
Tools work per project. If the agent asks which project, give it the project ID: the part of the project's address after /projects/.
To try it, ask the agent to build something ordinary, for example a delete confirmation modal, and to find existing components and tokens first.
See what agents did
- Account settings → AI agents → Connected agents lists every agent that called FISYCO with your account in the last 30 days: how it signed in, how many calls, the last call.
- Authorized agents on the same page lists agents you allowed through the browser. Revoke signs one out right away.
- In the project, Design system in code → AI agents shows a report for the last 30 days: developers with agents, existing components suggested, deprecated components steered away from, notes in pull requests and Missing components agents searched for and didn't find. Workspace owners get the same report by email every month.
FISYCO records which tool was called, by which agent, and the outcome. It doesn't store your source code.