Connecting an AI assistant to your ISMS safely (MCP access)

AI assistant access lets you use your own AI assistant, such as Claude, with your ISMS. Instead of copying ISMS content into a chat, you connect the assistant to ClearIMS through MCP (Model Context Protocol), an open standard for connecting AI assistants to the tools people use. Then you can ask "which risks are due for review this month?" or "log an incident: a phishing email was reported by Finance" in the chat you already have open.

The assistant works under the same rules as you do in the app.

Why it matters

Connecting an AI assistant is a new kind of access to your information, so the ISO 27001 controls that apply to any other access apply here too:

• Annex A 5.15 and 8.3 (access control and information access restriction): the assistant should reach only what the person using it is allowed to.
• Annex A 5.17 and 8.5 (authentication): access uses a credential that can be controlled, expired and revoked.
• Annex A 8.15 (logging): what the assistant does is recorded.
• Annex A 5.10 (acceptable use): your policies should say how staff may use AI with company information.

Connecting properly is safer than the alternative: staff pasting confidential ISMS content into an AI tool nobody has assessed.

How ClearIMS controls it

• Off until an admin turns it on. Your organisation decides whether to allow it.
• Personal access tokens. Each person creates their own named token on the AI assistant access page. The token is shown once, so store it safely.
• Read-only by default. A read-and-write token is optional, for everyday actions such as logging an incident.
• Same rules as the app. The assistant sees only your own organisation, and anything it writes is checked against your role. For example, only admins and analysts can log incidents.
• Approvals stay with people. Approvals, sign-offs, deletions and settings are never available to an assistant.
• Logged and limited. Every call is recorded in the audit trail, changes are marked "via AI assistant (MCP)", and requests are rate-limited.
• Revocable. Tokens expire, and you can revoke one at any time; it stops working immediately. Two-factor authentication is required.

What an auditor looks for

• A decision to allow AI access, and who it's allowed for.
• Least privilege: read-only by default, write access only where needed.
• Credentials are managed: tokens are named, expire and are revoked when people leave.
• An audit trail of what assistants did.
• A policy telling staff how they may use AI with company information.

Common mistakes, and how to keep it manageable

• Giving everyone a read-and-write token. Start read-only and add write access when there's a reason.
• Sharing tokens. Each person should have their own, so the audit trail shows who did what.
• Forgetting leavers. Revoke tokens as part of your leaver process.
• No AI use policy. A short section in your acceptable use policy is enough.

How to do it in ClearIMS

1. An admin turns on AI assistant access for the organisation.
2. Open AI assistant access and create a personal access token: read-only, or read and write if you need it. Copy it straight away; it's shown once.
3. Add ClearIMS to your assistant. See the connection guide for the details.
4. Review tokens regularly, and revoke any that are no longer needed.

What works today: assistants that accept an MCP server address and an access token, such as Claude Code. Support for more assistants, including ChatGPT, is planned. AI assistant access is available on all paid plans.

Related pages

• Connect your AI assistant: step-by-step connection.
• Account security: two-factor authentication.
• Audit trail: where assistant activity is recorded.
• Ask ClearIMS: the assistant built into ClearIMS.