We didn’t put AI in OneHQ. We put OneHQ where your AI already is.
Looking for AI in your practice management? Here’s the honest version. OneHQ has no AI model of its own — on purpose. It speaks MCP, the open standard that lets the assistant you already trust, like Claude or ChatGPT, reach into your practice through your own permissions. You choose the AI. OneHQ stays the source of truth it asks.
Illustrative example with invented names. Every read and every action is scoped to what the signed-in staff member can see and do.
“AI inside” vs an MCP licence
Plenty of software now advertises AI baked in. That means a model has been given a door into your client data, and you’re trusting the vendor’s judgement about what it can reach and what it does with it. OneHQ takes the opposite path — and for a system that holds tax files and identity records, we think it’s the only responsible one.
- A model you didn’t choose, with a door into your data
- When it’s wrong, it’s wrong inside your system of record
- You inherit the vendor’s answer on privacy and training
- Locked to one AI — you move when the vendor does
- You bring the assistant you already vet — Claude, ChatGPT
- OneHQ stays authoritative and audited — it decides what’s true
- Every request runs through your permissions, every time
- An open standard — swap or add assistants as they improve
Run your day by asking
OneHQ’s MCP server is live for Claude and ChatGPT. It’s a set of tools your assistant can call against your real practice — each gated by the same permissions your screens use. Most of what you’ll do with it is run your day, out loud:
Brief me on my day
“What have I got on, and who did I speak to?” Your meetings, calls and what’s come in, pulled together without opening a thing.
Make a note on a file
“Note that they’re switching banks.” Logged straight onto the client’s record, in your words — and it can be superseded later, never silently lost.
What’s on my list
Your tasks, your jobs and the next actions across the clients you look after — the list you’d otherwise assemble from three screens.
Move it along
Set a job’s WIP status, add a task, set a due date, reassign a job. The work advances while you talk — every change audited.
It also keeps an eye on what’s out for signature — so “what’s still unsigned?” is a question, not a hunt. And when there’s real accounting to do, it can do that too — safely:
Why MCP is the safe way to use AI on a practice
The MCP server is deliberately narrow. Every request runs through a system that decides — each time — what is true, what this person may see, and what they may do. The assistant asks; OneHQ answers.
- It sees what you see. Every call runs through the same access checks as your screens. A client you can’t open, the assistant can’t read — and it’s told an empty result isn’t proof the client doesn’t exist.
- It can’t post to accounts. Nothing is posted to a client’s Xero on the assistant’s word. Reclassifications land in their Xero as draft journals an accountant reviews and posts.
- It can’t contact your clients on a whim. The only outbound channel is SMS, and each text waits for a person to approve it in OneHQ. Email and document dispatch are refused entirely.
- A scope is a ceiling, not a grant. Connecting grants an assistant only what that staff member could already do — never more. Figures come back with where they came from, so an answer you didn’t compute is one you can still check.
Frequently asked
Does OneHQ have its own AI?
What is MCP?
Can the AI change my clients’ accounts?
Why not just build AI into OneHQ?
Is our client data used to train AI?
Bring your AI. We’ll bring the practice.
OneHQ unifies Xero, comms, jobs and compliance into one client record — and opens it to the assistant you already use, safely.
Start now