Connecting AI Clients via MCP - Documentation

Connect Meku to the AI tools you already use, such as Claude, ChatGPT, Cursor, and VS Code. Once connected, you can create, edit, and publish Meku projects by sending requests from that tool's chat.

The connection uses the Model Context Protocol (MCP), an open standard that lets AI tools work with other apps. You add Meku's server address to your AI tool, sign in to Meku, and approve access.

What you can do

After you connect, your AI tool can take these actions in your Meku account:

List your workspaces and projects.
Create a new project from a prompt or a template.
Send change requests to an existing project. The Meku agent edits the code and rebuilds the site.
Read project files and review the changes the agent made.
Update a project's project knowledge.
Enable a Supabase database for a project and run queries on it.
Remix a project into a workspace.
Publish a project to a meku.app address.
View visitor analytics for a published project.

Example requests:

“Which Meku workspaces do I have?”
“Create a Meku project for a bakery website with a menu page and a contact form.”
“In my bakery project, add an opening hours section to the home page.”
“Publish my bakery project.”

Server address

Use this address when you add Meku to any AI tool:

https://backend.meku.dev/api/mcp

Connect your AI tool

Follow the steps for your tool. Each guide uses browser sign-in. Where a tool also supports a personal access token, the token setup follows the main steps.

Claude (claude.ai and desktop app)

Open Settings > Connectors.
Click Add custom connector.
Paste the server address: https://backend.meku.dev/api/mcp
Sign in to Meku when prompted and click Allow.

Claude Code

Run this command in your terminal:

claude mcp add --transport http meku https://backend.meku.dev/api/mcp

Then run /mcp in Claude Code and select Authenticate. Your browser opens so you can sign in to Meku.

To use a personal access token instead, add it as a header:

claude mcp add --transport http meku https://backend.meku.dev/api/mcp \
  --header "Authorization: Bearer YOUR_PERSONAL_TOKEN"

Cursor

Open .cursor/mcp.json in your project. To make Meku available in all projects, use ~/.cursor/mcp.json instead.
Add the Meku server:

{
  "mcpServers": {
    "meku": {
      "url": "https://backend.meku.dev/api/mcp"
    }
  }
}

Open Cursor Settings > MCP. You can find it under Customize in the sidebar.
Turn on meku.
Click the sign-in prompt, sign in to Meku in your browser, and click Allow.

To use a personal access token instead, add a headers entry:

{
  "mcpServers": {
    "meku": {
      "url": "https://backend.meku.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_PERSONAL_TOKEN"
      }
    }
  }
}

Note: Cursor may still ask you to sign in through the browser when a token header is present.

ChatGPT and Codex desktop app

Open Customize > Plugins.
Click Add > Add custom MCP server.
Enter these settings:
- Name: Meku
- Connection: Server URL
- URL: https://backend.meku.dev/api/mcp
- Authentication: OAuth. Leave the advanced OAuth settings unchanged.
Select I understand and want to continue, then click Create as a plugin.
Click Connect Meku, sign in to Meku, and approve the access request.

VS Code

Open the Command Palette with Ctrl+Shift+P on Windows and Linux, or Cmd+Shift+P on Mac.
Run MCP: Add Server, select HTTP, and enter the server address. You can also add the server directly to .vscode/mcp.json:

{
  "servers": {
    "meku": {
      "type": "http",
      "url": "https://backend.meku.dev/api/mcp"
    }
  }
}

Click Start above the meku entry in the file.
Sign in to Meku in your browser and approve access.
Switch the chat to Agent mode. Meku tools only work in Agent mode.

To use a personal access token instead, add a headers entry:

{
  "servers": {
    "meku": {
      "type": "http",
      "url": "https://backend.meku.dev/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_PERSONAL_TOKEN"
      }
    }
  }
}

Test the connection

Ask your AI tool: “Which Meku workspaces do I have?” If the connection works, it lists your Meku workspaces.

Choose a sign-in method

Browser sign-in

Use browser sign-in for Claude, ChatGPT, Cursor, and VS Code. When you connect, the AI tool opens your browser. You sign in to Meku and click Allow. You never have to copy or store a secret key.

Personal access token

Use a token where a browser cannot open, such as a remote server or an automated script. A token is a secret key, so treat it like a password.

To create a token:

Go to Account > AI Clients (MCP).
In the Personal Tokens section, click Create Token.
Enter a name that shows where you use the token, such as Claude Code - Work Laptop.
Choose when the token expires: 7 days, 30 days, 90 days, 1 year, or Never.
Copy the token. It starts with meku_pat_. Meku shows the full token only once.

Note: Store tokens in a password manager or an environment variable. Do not commit them to a code repository. When a token expires, create a new one and update your AI tool.

Credits and models

Requests from an AI tool use the same credits and plan limits as work in the Meku editor. Creating a project and sending change requests both use credits.

The agent uses your workspace's default model unless you ask for a specific one. To compare models, ask your AI tool to list Meku's models. The list shows each model's credit cost and which models your plan includes. See LLM Switcher and Credits & Usage for details.

Approvals and secure inputs

Some requests pause until you make a decision.

Plan Mode: Ask your AI tool to use Plan Mode for larger changes. The agent proposes a plan and waits. Review the plan in your AI tool, then approve it, edit it, or reject it. See Plan Mode.
Secure inputs: Steps that need sensitive values, such as Stripe keys, can only be completed in the Meku editor. Open the project in Meku to finish them. See Secrets.

A new request to the same project cancels any pending pause. Answer the pause first if you want the paused work to continue.

Before you make changes

Database changes are permanent. Queries that add, change, or delete data affect your live database. Ask your AI tool to read the data first and to check with you before it runs a change.
Project Knowledge is replaced, not merged. An update overwrites the existing instructions. Ask your AI tool to read the current Project Knowledge before it updates it. The limit is 10,000 characters.
New projects are public by default. Ask for a private project if you need one. Private projects count toward your plan's private project limit.
Publishing makes your site live. Anyone can visit it at its meku.app address. Renaming the subdomain requires a paid plan.
Remixed projects start with a new chat history. The code is copied. Project Knowledge is copied only if you ask for it. You can remix your own projects and other users' public projects. See Remix.

Manage and revoke access

You can review and remove every connection from your Meku account.

Go to Account > AI Clients (MCP).
Check Connected Apps for browser sign-in connections and Personal Tokens for tokens.
Click Revoke next to any connection or token you no longer use or think is compromised.

Meku rejects requests from a revoked connection right away. To use that AI tool again, connect it again.

Troubleshooting

The AI tool cannot connect: Check that the server address is exactly https://backend.meku.dev/api/mcp.
Cursor shows a connection error: Open the Output panel and select MCP Logs to see the error.
Meku tools do not appear in VS Code: Switch the chat to Agent mode.
Requests stop working after a while: Your token may have expired or been revoked. Check Account > AI Clients (MCP) and create a new token if needed.
A request stays paused: The agent is waiting for an approval or a secure input. See Approvals and secure inputs above.

For other issues, see Troubleshooting & Support.