MCP server

The server speaks the Model Context Protocol (MCP), the standard AI coding tools use to reach outside data. It reads the live registry, so new components show up within minutes with no update. Every tool only reads, except disconnect, which ends your own agent's connection.

Connect with sign-in

Add this address to your agent. It speaks Streamable HTTP. Sign-in is in beta.

https://mcp.aicanvas.me

1. Add the address to your agent.
2. Your browser opens when the agent connects, or when you click its Connect button (in Cursor: Settings, Tools & MCPs). Sign in to AI Canvas or create a free account, then click Allow right away: some agents stop waiting in less than a minute.
3. Ask for what you want in plain words.

Your account stays yours

Your agent gets its own sign-in. It never gets your password or your browser session. It cannot change your account, your password or your plan, and it cannot buy anything.

Each sign-in lasts an hour and renews itself until you disconnect the agent. You get an email when a new agent connects.

See and disconnect every agent in Settings, Connected AI agents. An agent can also disconnect itself with the disconnect tool.

No browser? CI and remote shells

Send your API key to the same address as the header Authorization: Bearer <your key>. Get or rotate the key in Account settings. Store it as a secret in your CI, never in git. Use either the browser sign-in or the header, never both: a saved header overrides the sign-in, so every call would fail.

claude mcp add --transport http --scope user aicanvas https://mcp.aicanvas.me --header "Authorization: Bearer aic_your_key"

Stuck signed out?

AI Canvas tools may stop answering, or your agent may say it is not authorized. Possible causes:

• You disconnected the agent in settings.
• The sign-in took longer than your agent waits. Some wait less than a minute. If the browser says it cannot reach localhost after you click Allow, your agent had already stopped waiting.
• The browser says “This sign-in request has ended” because the request expired or was already answered.
• The browser says “Your AI agent needs to connect again”: its saved connection is no longer valid. Starting again reuses it, so remove the server from your agent and add it again.
• If you use the header method, you rotated your API key.

To fix it, start the sign-in again from your agent. For Claude Code, run claude mcp login aicanvas. For Codex, run codex mcp login aicanvas. In other agents, use the reconnect or sign-in action, or remove the server and add it again. If you are already signed in to aicanvas.me in that browser, it goes straight to Allow. For the header method, copy your new key.

In Cursor, click Authenticate when it appears in the chat, or open Cursor Settings, Tools & MCPs and click Connect next to aicanvas. Until you sign in again it shows Error there: click Logout, then Connect. If the Error stays after that, quit Cursor and open it again.

VS Code asks to sign in again by itself: click Allow, then Allow in the browser. If Copilot answers without AI Canvas, start your prompt with #aicanvas.

Codex does not open the sign-in by itself: its AI Canvas tools answer “Authentication required”. Run codex mcp login aicanvas in a terminal, then click Allow.

In the ChatGPT desktop app, open Settings, Plugins, MCPs and click Authenticate next to aicanvas. ChatGPT and Codex share that list, so signing in from one signs in both.

What we store

For each connected agent, we store its name, the address it returns to after sign-in, when it connected, when it was last used, and the country of the browser and the agent. Sign-in tokens are stored only as hashes. Disconnecting deletes the connection. Read the Privacy Policy.

Run it locally with npm

Pick your agent. Replace your-token with the token from Account settings, or sign in on the MCP page and copy a command with your token already in it.

claude mcp add aicanvas --scope user --env AICANVAS_TOKEN=your-token -- npx -y @aicanvas/mcp

codex mcp add aicanvas --env AICANVAS_TOKEN=your-token -- npx -y @aicanvas/mcp

{
  "servers": {
    "aicanvas": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@aicanvas/mcp"],
      "env": { "AICANVAS_TOKEN": "your-token" }
    }
  }
}

{
  "mcpServers": {
    "aicanvas": {
      "command": "npx",
      "args": ["-y", "@aicanvas/mcp"],
      "env": { "AICANVAS_TOKEN": "your-token" }
    }
  }
}

{
  "mcpServers": {
    "aicanvas": {
      "command": "npx",
      "args": ["-y", "@aicanvas/mcp"],
      "env": { "AICANVAS_TOKEN": "your-token" }
    }
  }
}

{
  "mcpServers": {
    "aicanvas": {
      "command": "npx",
      "args": ["-y", "@aicanvas/mcp"],
      "env": { "AICANVAS_TOKEN": "your-token" }
    }
  }
}

Try it

Start every prompt with “use the aicanvas MCP”. Agents often skip MCP tools and search the web unless you name them.

Use the aicanvas MCP to list the component categories.

It should list the categories with their counts. Then try:

• "Use the aicanvas MCP to find a navbar and install it."
• "Use the aicanvas MCP to plan a settings page from Andromeda Pro, then build it."
• "Use the aicanvas MCP to check the components in this file against their props."

Tools

Find

Read

Plan and check

Install

Icons

Account (sign-in only)

The best order for a new page: compose_page to plan, get_component_props before writing each component, validate_usage on the finished file, get_audit_checklist before calling it done.

Accounts and Premium

• Browsing, search, props and install commands need no account.
• Reading source needs a free account: set AICANVAS_TOKEN.
• Premium components, Andromeda Pro, its templates and the AI Brain need an account with Premium.
• Icons in the Line style install with a free account. Duotone, Solid and Solid duotone need Premium.
• With sign-in, your plan comes along: Premium items show as locked with an upgrade link, and upgrading unlocks them on the same connection. There is no daily request limit.

Without a token, a source request returns a short note asking for a free account instead of the code.

Configuration

npx can reuse a copy it cached earlier. To always start the newest version, use @aicanvas/mcp@latest in the command. The package is on npm.

Troubleshooting

The tools do not show up

Restart your tool after adding the server; most read their MCP list only at start. Check that Node.js is installed and that npx runs in a terminal.

"Failed to spawn process" or a network error

npx could not fetch the package. Check your connection or proxy, then restart the tool.

Source comes back as a sign-up note

The token is missing or wrong. Copy it again from Account settings and update the AICANVAS_TOKEN value in your config.

Test the server on its own

This opens a browser window where you can call each tool by hand:

npx -y @modelcontextprotocol/inspector npx -y @aicanvas/mcp