Kitsuvo MCP server and plugins for Claude Code, ChatGPT and Codex

The Kitsuvo plugin gives an AI client a set of tools for driving the Kitsuvo browser on your own computer. The client can open a page, read it, and ask the browser for its detection report, which lists the AI-builder fingerprints, slop signals and impersonation findings behind a score. It works with Claude Code and with ChatGPT or Codex hosts that support local MCP servers.

It is the same detector described on the home page, exposed over the Model Context Protocol (MCP). Nothing here is needed to browse with Kitsuvo; it is for people who want an AI client to inspect sites for them.

What the plugin does

The plugin bundles Kitsuvo's native MCP server with four skills for setup, browsing, detection evidence and responsive checks. The server exposes:

• Navigation and tab management, accessibility snapshots, clicks, typing, hovering, selection, key presses and waits.
• Page text, JavaScript evaluation and responsive sizing.
• kitsuvo_report, which returns the AI-built, slop and impersonation evidence for a page and describes hidden or labelled sites on search results.
• Screenshots, on Windows only.

No API key is needed for local browser automation or local detection.

Install in Claude Code

1. Install the browser. Install Kitsuvo on Windows or macOS and make kitsuvo available on PATH. The plugin does not install the native browser.
2. Add the marketplace. The integration marketplace is a public GitHub repository.

claude plugin marketplace add leancoderkavy/kitsuvo-integrations
claude plugin install kitsuvo@kitsuvo-browser

3. Run the setup skill, then ask: Open https://example.com in Kitsuvo and explain its detection evidence. Tool names may carry a host-specific prefix.

To add only the MCP server, without the skills, use claude mcp add kitsuvo -- kitsuvo mcp. If kitsuvo is not on PATH, replace the command with the absolute path to the executable and keep the mcp argument.

Install in Codex or the ChatGPT desktop app

codex plugin marketplace add leancoderkavy/kitsuvo-integrations
codex plugin add kitsuvo --marketplace kitsuvo-browser

In the ChatGPT desktop app, open the Plugins directory, select Kitsuvo Browser and install Kitsuvo. Restart the host after changing packages. Local execution needs a host with local MCP support. A web or cloud conversation needs a connection to the computer that runs the browser, such as a secure tunnel to the local HTTP bridge or the opt-in cloud relay described in the integrations guide.

You can also download the integrations archive and install from the extracted folder.

Open the browser inside ChatGPT or Claude

For web chat, use the cloud connector in App mode. A desktop plugin alone does not connect web chat or add its browser panel. Kitsuvo runs the pages on your computer; your AI app shows the panel.

1. Start the connector. Install Kitsuvo and Node.js 22 or newer, extract the integrations archive, then follow the cloud setup guide to start App mode. Type CONNECT when the terminal asks. Keep it running.
2. Add the connection in your AI app. Use https://kitsuvo-mcp-relay.fly.dev/mcp with OAuth. In ChatGPT, add a custom MCP server from Plugins. In Claude, open Customize → Connectors and add a custom connector; choose “Register automatically” for its OAuth client. Team and Enterprise owners add it through Organization settings → Connectors first. Your account or workspace must allow custom connections.
3. Approve your browser. The authorization page gives a one-time request code. In the connector terminal, enter approve <code>, review the client and return address, then enter APPROVE. Refresh the authorization page to finish. Approve only the client you just added.
4. Open a new chat. Enable or select the Kitsuvo connection, then send: Open Kitsuvo in chat at https://example.com, then explain its detection evidence.

Success is a browser panel in the conversation. Windows includes page previews; macOS uses page text and controls. The connector shares requested page text, screenshots, tool arguments and reports through the relay with your AI app. Closing it disconnects access. ChatGPT connection instructions and Claude interactive connector instructions describe the host-side steps. A workspace Secure MCP Tunnel is an alternative for ChatGPT; see the full integration guide.

What the AI client can see

• Page content goes to your client. Page text, screenshots, tool arguments and reports returned through MCP are visible to the connected AI client and covered by that client's data handling. Kitsuvo's local-only switch does not stop a client from receiving tool results.
• A separate profile. Automation uses its own browser profile, and it refuses sensitive sign-in pages and password fields by default. The packages do not enable the main profile or the sensitive-page override.
• Local by default. The plugin adds no analytics. Adding it does not turn on cloud verdict engines.

The full account of what Kitsuvo handles is in the privacy policy.

Status and limits

• The plugin works from the public GitHub marketplace above. It has not been approved in Claude's or OpenAI's public plugin directories, so it does not appear there yet.
• It needs the desktop browser. It does not install it, and it does not work without it.
• One bridge connects to one agent profile, and simultaneous callers share its tabs.
• An AI-built verdict is provenance, not proof of harm, and a low score is not proof of human authorship or safety.

How it differs from a web checker's MCP tool

Some web checkers also expose an MCP tool. Is That Vibe Coded, for example, offers one that fetches a URL on its servers and returns a score. The Kitsuvo plugin is the other shape: it drives a browser on your computer, so it can read pages as they render, work through search results and run responsive checks. For a side-by-side of the web checkers, see AI website detector alternatives.