AI connectors
Connect the system you already run to Claude and ChatGPT.
An MCP connector lets an AI assistant read and act inside your management system as the signed-in user. Your staff ask in plain language; the system does the work, within their permissions, on the record.
What MCP is, in one paragraph
The Model Context Protocol (MCP) is the open standard that Claude, ChatGPT and other assistants use to talk to outside systems. A connector is a small, secure endpoint on your own server that describes the things the assistant may do ("list arrears", "queue a reminder", "approve a requisition") and performs them when asked, as the person who signed in. The assistant never gets a database password and never sees more than that person would.
What it changes for your staff
- Questions in plain language. "Who has not paid this module?" instead of six clicks and a filter.
- Routine work by chat. Reminders, approvals, lookups and reports from a phone, in the chat app staff already use.
- Managers get answers, not spreadsheets. A head of department asks for the week's figures and gets them, with the source.
- Nothing leaves your system. Data is read on request, in your system, under your permissions and audit trail.
What I build
- A connector on top of your existing system: PHP, Laravel, WordPress, Odoo, or a custom build in another stack.
- A tool set designed around your roles: finance, registrar, HR, pastor, manager. Each tool is scoped to what that role may do.
- Sign-in through your system's own accounts (OAuth), with two-step sign-in for administrators.
- An audit trail of every call, and the security checks a normal code review misses, closed before go-live.
- Set-up guides for Claude, ChatGPT and Claude Code, and a short training for staff.
Already running
Connectors are in daily use on a college system, a church platform and an author's writing studio, each with dozens of role-scoped tools. Ask and I will show one working.
registrar › Admit the 12 applicants I approved this morning into Diploma ICT, intake Jan 2027. assistant › Found 12 approved applicants. Admission numbers will be ADM0501 to ADM0512. Confirm? registrar › Confirm. assistant › 12 students admitted. Fee invoices for module 1 created. Welcome SMS queued. // tools: admissions.list, admissions.admit, fees.invoice, sms.queue // signed in as: registrar (two-step verified) · logged
Audit
I read your system and list the tasks staff repeat most.
Design the tools
A short, role-scoped tool set, agreed with you in writing.
Build and test
Connector, sign-in, audit trail and security review on a copy of your data.
Connect and train
Staff connect Claude or ChatGPT in minutes; I train the first users.
Straight answers
Questions about AI connectors.
Does the assistant get access to my whole database?
No. It can only call the tools I build, each limited to what the signed-in user may do, and every call is logged. There is no direct database access and no bulk export.
Does my data go to the AI company?
Only what a user asks for in a given conversation is sent to the assistant, the same way it would be if the user copied it into a chat. Nothing is synchronised or stored with the AI provider by the connector.
Do I need to rebuild my system?
No. The connector sits beside your existing code and uses your existing accounts and permissions.
Which assistants work?
Claude (web, desktop and Claude Code) and ChatGPT today; any assistant that supports MCP in future.
How long does it take?
A first useful connector with a focused tool set typically takes two to four weeks, including the security review.
Want your system to answer when staff ask?
Tell me what system you run. I will say what a first connector should do and what it costs.