Kitana is an MCP server. An agent connected to it can run a brand review, read the findings and compare runs without opening the web app. The server URL is the same for every client:

https://mcp.replikit.ai

Before you connect

You need RepliKit Pro. A workspace whose plan does not include MCP is refused when you approve the connection and on every tool call.

Sign-in uses OAuth 2.1. The first call returns 401, your client finds the authorization server, and you approve the connection in a browser. There is no key or client ID to set up by hand.

run_review spends one credit per review of up to 10 assets, the same as a review on the web. The other tools are free.

Connect Claude

In claude.ai or Claude Desktop, open Settings → Connectors → Add custom connector, and paste the server URL.

https://mcp.replikit.ai

Approve the connection in the browser window that opens.

Connect Claude Code

Add the server from your terminal:

claude mcp add --transport http kitana https://mcp.replikit.ai

Then run /mcp in Claude Code, choose kitana and sign in.

Connect Cursor

Add the server to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "kitana": { "url": "https://mcp.replikit.ai" }
  }
}

Cursor registers itself with the authorization server, then asks you to approve the connection in a browser.

Connect ChatGPT

Open Settings → Apps & Connectors, turn on developer mode, then choose Create. Paste the server URL and pick OAuth for authentication.

https://mcp.replikit.ai

To review an image, attach it with the paperclip. Ask for a review with nothing attached and the upload widget opens instead.

Connect another client

Any client that supports remote MCP servers over HTTP with OAuth can connect. Add https://mcp.replikit.ai as a remote HTTP server and leave authentication to OAuth. The client discovers the rest.

The tools

run_review: Reviews 1 to 10 images against your brand kit and spends one credit.
prepare_upload: Reserves an upload for local images and returns signed URLs with ready-to-run curl lines, for agents with a shell. It spends nothing.
get_review: Returns one review in full, by its ID, including a review that is still running.
list_reviews: Lists the reviews in your workspace, newest first, up to 25 at a time.
compare_reviews: Compares two reviews and returns the change in each category, whether they are comparable, and any caveats.
switch_kit: Lists your brand kits and selects the one this connection reviews against. It changes nothing on the web or in Slack.

Images can be PNG, JPEG, WebP or GIF. SVG is not accepted.

What a review returns

A finished review carries report, a preformatted text report drawn for a monospace terminal. Show it to the user exactly as given. It sits next to the overall score, the verdict, the five category scores, the findings and a link to the report in Kitana.

run_review waits up to two minutes. A review that takes longer comes back with status running; call get_review with the same ID for the result. Checking is free.

In Claude and ChatGPT, a review widget also appears in the chat.