AI assistant (MCP)
Drive AudiScale from your own AI assistant.
Claude, ChatGPT, Cursor, VS Code: plug the assistant you already use into AudiScale’s MCP server. It reads your audits, action plan and fixes, exactly as the interface shows them. Anything it asks to change becomes a proposal you approve.
7-day trial, cancel before you’re charged.
- Claude, ChatGPT, Cursor, VS Code
- The same data as the interface
- Every change goes through you
Connect
One address to add, one authorisation to give
On claude.ai, Claude Desktop, ChatGPT and tools that accept a remote MCP connector, you add the AudiScale server address as a custom connector, then authorise the app on the screen that opens: no key to copy.
For Claude Code, Cursor and VS Code, AudiScale provides the command or the button that adds the server. A dedicated access key remains available for any compatible client. You see the active accesses on your account and remove them whenever you like.
- No key to copyA custom connector and an authorisation inside the assistant.
- For coding toolsClaude Code, Cursor, VS Code with GitHub Copilot.
- Revocable accessEach access is managed and removed from your settings.
What it sees
Your projects, as the interface shows them
Your assistant reaches what you see in AudiScale: audits, action plan, radar, connectors and fixes, project by project. You ask in your own words; it looks up the answer in your measured data.
For the inbox, it works at account level and never receives the body of an email.
- Audits
- Action plan
- Radar
- Connectors
- Fixes
- InboxAt account level, never an email body
Proposed, not executed
Every change stays a proposal for you to approve
When your assistant asks for a change, AudiScale files a proposal and returns its reference. Nothing is published, edited or sent until you approve it, and the assistant needs your explicit agreement for each proposal, never in bulk.
If you turned on autopilot, a proposal follows the level you chose: it may go on its own when its veto window ends, unless you stop it. A page or an article always waits for your design approval on a real preview.
- 1Your assistant asks
- 2AudiScale files a proposal
- 3You approve
- 4The action runs, logged
Nothing moves before you agree
- One proposal, one decision. Your assistant proposes; you approve, in AudiScale or in the conversation.
- Design approved on a preview. Pages and articles go through the real preview before going live.
- Tracking and log. The assistant follows an approved action as it runs; the history stays in AudiScale.
For your developer
Code fixes, applied where the code lives
For a site that is not on WordPress, your developer’s assistant reads AudiScale’s prioritised fixes, with the target pages, the steps and the expected result, then applies the changes in the code itself. AudiScale never touches your repository; the next measurement checks the result.
From a fix, a button copies the request ready to paste into the assistant: it holds no key, the assistant reads the request itself through the AudiScale MCP.
Which assistants are compatible?
Claude Code, Claude Desktop and claude.ai, ChatGPT, Cursor, GitHub Copilot in Visual Studio Code, and any MCP client that accepts a remote server.
Can my assistant change my site without me?
No. Any change it asks for becomes a proposal, which only acts after your approval. The one exception is yours to choose: with autopilot turned on, a proposal follows your autonomy level, with veto and undo.
Do I need a specific plan?
Connecting your AI assistant is included in all three plans: Essentiel, Pro and Max.
Does my assistant read my emails?
It can check the state of your inbox, at account level, but never receives the body of an email.
Watch it work on your site
Enter your address: the 7-day trial starts, and the first audit runs as soon as you connect.
Start the trial7-day trial, cancel before you’re charged.
