Overview
Your team has already picked an AI assistant. It is open in a tab right now, it has your context in it, and nobody wants to learn a second one just to change a headline on the website.
So connect the assistant you already use to your website through our governed MCP server — with the same permissions, approvals and audit trail you get in the portal. Ask Claude or ChatGPT to update your opening hours and the change flows through exactly the same guardrails as if you had asked here.
MCP — the Model Context Protocol — is the open standard for giving AI clients safe, scoped access to external systems. We expose your site through it as a governed server rather than as a raw API key, which is the difference between delegating a task and handing over the keys.
How it works
- Add the AWT MCP server to your AI client and authorise it as yourself.
- Ask for a change in that client, in plain language, as you normally would.
- Your permissions are checked before anything is touched.
- The change is made as a draft or on staging, with a preview where relevant.
- It is recorded in the audit log, marked “via MCP”, and waits for approval to go live.
What you can do from your own client
- Read and query your content — what is published, what is stale, what is missing metadata.
- Draft and edit pages and posts, with the same field validation as the portal.
- Request publishing — subject to your role and any approval rules in place.
- Pull reporting figures into a document or deck you are already writing.
- Combine your site with your other tools in one workflow — the point of a shared protocol.
Why teams use it
- Any MCP client — ChatGPT, Claude, Copilot and more.
- Same guardrails — permissions, approvals and audit apply everywhere, with no exceptions.
- Metered and revocable — see usage per client and revoke access at any time.
- No new habits — your team works where they already work, so adoption is immediate.
Staying in control
Every change follows the same rules regardless of where it was requested: it lands on staging or as a draft, it is scoped to the permissions of the person who asked, and it waits for approval before going live. Connecting an external client grants no extra power — an Editor in the portal is an Editor in Claude.
Access is per person and revocable in one click, and every MCP-originated action is tagged in the audit log so you can always see which changes came from outside the portal and who authorised them.
Availability
Use your own AI (MCP access) is included on Scale and Enterprise plans.
Frequently asked questions
Which AI clients are supported?
Any client that speaks the Model Context Protocol, which includes Claude, ChatGPT and GitHub Copilot, plus a growing list of editors and custom internal tools. Because MCP is an open standard, new clients work without us having to build a specific integration.
Is this less safe than using the portal?
No — the guardrails live on our side, not in the client. Permissions, draft-first behaviour, approval rules and the audit log all apply identically, so an external assistant cannot do anything you could not do in the portal yourself.
Does our content get used to train someone else's model?
Not by us. What the AI provider you connect does with the conversation is governed by your agreement with them, which is worth checking if you are in a regulated sector — most business and enterprise tiers exclude training by default.
Can we see what an external client has been doing?
Yes. Every action taken through MCP is tagged as such in the audit log with the person who authorised the client, and usage is metered per connection so you can see volume at a glance and revoke anything that looks wrong.
How do we turn it off?
Revoke the connection in Settings and access stops immediately — for one person or for the whole workspace. Any changes already made remain in the audit log, and nothing in your site depends on the connection staying open.
