AI Assistants (MCP)

Connect an AI assistant such as Claude or ChatGPT directly to your billing platform. A connected assistant can answer reporting questions in plain English, look up records, and - where you deliberately allow it - carry out platform actions on your behalf.

The connection uses the Model Context Protocol (MCP), an open standard supported by Claude, ChatGPT, and a growing number of other assistants.

What a Connected Assistant Can Do

You choose what each connection is allowed to do when you approve it. There are three levels of access, and each is also controlled by permission flags on your own user account:

Reporting KPIs (read-only aggregates)

The assistant can answer dashboard-style questions from the platform’s KPI data:

Revenue and invoicing by month, quarter, or year
Payments received, by method and period
Direct Debit collection performance, including failures and retries
Outstanding debt aged by how overdue it is
Customer, number, and feature growth and churn

It only sees aggregate figures at this level, never individual records.

Individual records (read)

The assistant can find and read individual customers, numbers, features, services, invoices, payments, transactions, notes, tickets, contacts, and TODOs. It can search customers from a name, account number, postcode, or phone number, and filter flexibly - including by parent-customer criteria such as “invoices for customers added in the last 12 months”.

Lists return compact summaries by default, with full record detail available on request.

Individual records (create, edit and actions)

Where you deliberately grant it, the assistant can create and update records and run the platform’s own workflow actions: suspending a number, emailing an invoice, raising a refund, writing off debt, and so on. Actions go through exactly the same workflows and validation as the platform interface.

Write access is never granted by default. It must be ticked separately when you approve the connection, and it needs the corresponding flag on your user account.

How Access Is Controlled

The connection is designed so that an assistant can never see or do more than the person who approved it:

Your own login: each member of staff approves their own connection through the normal platform login and a two-factor authentication code. No passwords or keys are ever shared with the assistant.
Your own permissions: everything the assistant can see is limited by your account’s own access rights, including customer scoping.
A platform-wide ceiling: your platform’s MCP settings cap what any connection can reach, with separate read and write kill switches for the whole platform.
Short-lived credentials: connections use rotating tokens (one-hour access tokens with a rolling sixty-day renewal), and replayed or stolen tokens are revoked automatically.
Full auditability: you receive an email whenever a connection is approved or revoked, and everything an assistant does is recorded in the activity log, flagged separately from API activity.
Revocable at any time: you can review and revoke your connections from the AI Assistants (MCP) page on the platform’s Advanced menu.

Getting Started

Initial setup - what you need, and how to connect Claude, ChatGPT, or another assistant
Using AI assistants - what to ask, and how to manage your connections

Want a billing platform your AI assistant can talk to? see how the SAFE Billing Platform connects to Claude, ChatGPT, and other AI assistants