AI Agent Access (MCP)

Violentmonkey can be managed by AI agents such as Claude through the Model Context Protocol (MCP). Instead of copying code back and forth, you can simply ask the agent to list, create, edit, or fix your userscripts, and it will do so directly.

Setup

1. Add the server to your agent

With Claude Code:

claude mcp add violentmonkey -- npx -y @violentmonkey/mcp -p 5678

With Claude Desktop or any stdio-based client, add this to the mcpServers config:

{
  "mcpServers": {
    "violentmonkey": {
      "command": "npx",
      "args": ["-y", "@violentmonkey/mcp", "-p", "5678"]
    }
  }
}

2. Connect Violentmonkey

Ask the agent for the Violentmonkey connect URL. It looks like this:

https://violentmonkey.github.io/mcp_connect.html?port=5678&token=<generated_token>

Open it in your browser. Violentmonkey intercepts the URL and takes you to the AI agent access (MCP) section in settings, where you can approve the request. After approval, the extension connects to the server and the agent can manage your scripts.

3. Use it

“List my userscripts”, “Create a script that hides the sidebar on example.com”, “Fix the bug in my GitHub script”.

Connecting from another device

By default the server runs in stdio mode: your agent starts it as a child process, and everything stays on your own machine. If your agent runs elsewhere — for example on a remote server or a mobile device — use the HTTP mode instead:

VM_MCP_TOKEN=secret npx -y @violentmonkey/mcp -p 5678 --transport http

Then register it with your agent:

claude mcp add --transport http violentmonkey http://127.0.0.1:5678/mcp --header "Authorization: Bearer secret"

A few things to know about this mode:

• The token acts as a password. The server generates a random one on each start, so set a fixed token with --token or the VM_MCP_TOKEN environment variable — otherwise the agent’s saved credentials stop working after a restart, and the extension can no longer connect either.
• The server only listens on your own machine (127.0.0.1) by default. If you open it up with --host 0.0.0.0, put it behind HTTPS and keep the token secret, like any password exposed to the network.

Notes

• The connection request shows the server’s port. Only approve it if you started the server yourself.
• The port and token are kept in memory only. After a browser restart, open the connect URL again to reconnect.
• Disconnect at any time from the same settings section.