Connect Creght to your AI assistant (MCP)

What Creght MCP is for

Once Creght is connected to the AI assistant you already use (Claude, Claude Code, Cursor and others), you can ask it to look after your website right in the chat, without opening the editor. For example:

Check your numbers: "How did my site do last week, and where did the traffic come from?"
Publish content: "Post this draft to the blog, using the image I just sent as the cover."
Handle inquiries: "Put this week's contact form submissions into a table for me."
Go live: "I've checked the preview, publish it."

How it differs from the Creght Skill: the Skill, together with the CLI, is for writing site code (adding pages, changing styles) and needs a coding environment that can run commands. MCP is for day-to-day operations (content, analytics, inquiries), and works in chat-only assistants such as Claude on the web.

How to connect

First check which site your account is on; accounts don't carry over between them:

creght.com accounts: https://creght.com/api/mcp
creght.cn accounts: https://creght.cn/api/mcp

Note that it ends in /api/mcp; without /api it won't connect. The steps below use creght.com.

Claude on the web or desktop

Open Settings → Connectors and click "Add custom connector".
Give it any name, such as Creght, set the URL to the address above, and leave everything else empty.
Click Connect. Your browser opens the Creght consent page; sign in and click Allow.

Claude Code

Run this in your terminal:

claude mcp add --transport http creght https://creght.com/api/mcp

Then type /mcp in Claude Code, pick creght, and authorize in the browser when prompted.

Cursor

Open MCP in Cursor's settings, add a new server, and enter:

{ "mcpServers": { "creght": { "url": "https://creght.com/api/mcp" } } }

Save, then authorize in the browser when prompted.

Any other assistant that supports remote MCP takes the same address. There is no key to enter.

What to check when authorizing

The consent page says "unverified". That's expected: the app name comes from the assistant itself and Creght can't confirm it. What you should check is the callback address shown on the page. It should be the assistant you're using (for example claude.ai, or 127.0.0.1 on your own machine). If you don't recognize it, don't click Allow.
Access comes in two levels: read and write. With read access only, the AI can see your analytics and content, but posting articles, uploading images and publishing the site will be refused.
The AI can reach exactly what your account can: your own projects, plus projects others have invited you to.
One authorization lasts a long time; if you use it regularly you won't need to sign in again.

Changes go live immediately

Please keep this in mind:

CMS content the AI creates or edits is live as soon as it's saved. There is no draft step.
When the AI publishes your site, every visitor sees the new version right away.

A safe habit is to add "show me the content before you write anything, and wait for my OK" to your request. Most assistants also ask before running a write action, so think twice before clicking "Always allow".

What it can't do

Edit site code, add pages or change styles: use the Creght Skill or the editor.
Delete CMS entries or change a CMS collection's fields: do that in the editor.
Upload files over 20 MB: use creght upload from the Creght CLI, or upload in the editor.

Summary

With Creght MCP connected, you can hand analytics, content and inquiries to your AI assistant. Remember two things: the address ends in /api/mcp, and write actions go live immediately. The full tool list and parameters (written for AI) are in Connect to Creght over MCP.