Use Claude or ChatGPT with OneHQ

What is OneHQ’s MCP server?

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to real systems through defined, permission-checked tools. OneHQ’s MCP server is a set of those tools: your assistant calls them to look things up in your practice and, where you allow it, to record what you have just done. Every call runs through the same access checks as OneHQ’s own screens. MCP licences are included in OneHQ’s price for everyone who signs in.

What is the OneHQ MCP server address?

https://app.onehq.com.au/api/mcp

It’s the same address for every practice and every person; OneHQ works out your practice from who signs in, never from the address. In OneHQ it’s shown with a copy button under Settings → Integrations → My MCP connection. Sign-in is OAuth, so there is no API key to create and nothing else to paste.

Which assistants work with OneHQ?

Claude and ChatGPT. OneHQ’s is a standard MCP server, so other assistants that can add a remote MCP server with OAuth sign-in use the same address. Whichever you use, add OneHQ from the assistant’s website in a web browser, not from its desktop or phone app: Claude’s desktop app shows the same screen but its sign-in never completes, and ChatGPT’s app doesn’t have the setting at all. Which plans of Claude or ChatGPT let you add your own MCP server is set by Anthropic and OpenAI.

Who can switch it on?

Three things must all be true before an assistant can reach OneHQ, and each belongs to a different person.

The practice switches it on

It starts off for every practice. Someone with the System settings permission (a Partner, by default) turns it on under Settings → Integrations. The same place holds two more firm-wide choices, both off until someone turns them on: whether assistants may text clients, and whether they may read the words in a client’s documents rather than only a document’s figures.

Each person holds the MCP permission

Each staff member needs the MCP permission in Users & Permissions. It’s on by default for every role, so a practice can switch it off for a particular person.

Each person decides for their own connections

Under My MCP connection, each person can keep their assistant from texting clients even where the firm allows it, and can choose to let it prepare a client’s year for review.

OneHQ itself can also pause every connection at once, if it ever needs to, without disconnecting anybody.

How do I connect Claude or ChatGPT to OneHQ?

A connection asks you again only when a new kind of access is involved, such as texting clients for the first time. One that hasn’t been used for 60 days needs reconnecting.

What can an assistant read in OneHQ?

Only what the signed-in person can already see in OneHQ. Within that:

• Clients and their history. Find a client by name, group or email address; their details, the staff responsible, their registrations, open jobs and group; and their timeline of calls, notes, meetings, emails, documents out for signature and job changes, with the record behind any one event, such as a voicemail transcript or meeting notes.
• The work. Jobs filtered by client, type, status, period, overdue or yours; what needs attention today; what’s out for signature; and AML/CTF matters with their status and risk rating.
• Meetings, calls and messages. The practice diary, prospects, a client’s meetings, phone calls and text messages, and your own calls and meetings nobody has written up yet.
• Documents. The files in a client’s folder; a document found by what it is (“her 2025 tax return”); what a document states, such as its type, period and figures; and the attachments on a client’s recent emails. Where the practice allows it, the words of a document too, with tax file, bank, card and Medicare numbers, licence and passport numbers and dates of birth masked. A scan is read with Amazon Textract, within a monthly page allowance.
• The accounts, for a client with a Xero file linked. Lodged BAS figures, profit and loss and balance sheet with the accounting basis stated, the chart of accounts, ledger transactions by supplier, narration or account, the year month by month (actual and forecast), year-end review status and suggested fixes, draft journals not yet posted, a company’s tax position and franking account, farm management deposits and livestock, and the client’s queries for the year.
• Your firm’s templates. The Word and Excel templates your firm keeps in Files, so a letter or schedule is made in your own format.

What can an assistant change in OneHQ?

A deliberately narrow set of things, each recorded under the person’s own name, exactly as if they had done it on screen:

• Notes, calls and meetings on a client’s timeline, or on a prospect’s. It can correct or archive a note or meeting; the earlier wording is kept and nothing is deleted.
• Tasks and jobs. Raise a client task for you or a colleague; change a job’s status, owner or due date; set its one-line WIP note.
• Client queries. Add, answer and tick off a client’s queries, and draft the email that puts them to the client. You send it from OneHQ.
• A new document in a client’s Files. Save an email attachment to the client’s folder, make the client’s copy of one of your firm’s templates, or save a workbook or document it drafted, named “(AI draft)”. It never changes or deletes a file already there.
• Draft journals. Reclassify transactions, move an account’s balance, write a journal or accept a suggested fix. It shows you the debits and credits first, and records them only when you confirm, as a draft journal in the client’s Xero file for an accountant to review and post. It can also map accounts and key year-end figures such as farm management deposits and livestock.
• A text message, staged for a person. Only where the practice and the person both allow it, and only for someone who holds the permission to text clients in OneHQ. The assistant shows the exact words and who they go to, then puts the text in OneHQ; a person opens it there and presses Send. It goes only to the mobile on the client’s record, never to anyone who replied STOP.
• Year-end preparation, for a person who turns it on for themselves. The assistant prepares a client’s year and leaves cards on OneHQ Review for that person to accept or dismiss. It never posts.

The one thing it can remove is a draft journal that hasn’t been posted, and only under an access you approve separately.

What can an assistant never do through OneHQ?

• Send an email, or send a document for signing.
• Post to a client’s Xero file, or change a transaction the client entered.
• Delete a client, a job, a note or a file.
• See a client you can’t open in OneHQ. As in OneHQ itself, a client’s name is practice-wide but the detail is not, and a restricted result is marked as restricted.
• Receive a tax file number. Every reply is passed through a mask that hides tax file numbers before it leaves OneHQ.
• See an AML matter under internal review. Those stay in OneHQ.
• Take instructions from a client. Words a client wrote (a message, an email, a transcript or a document) come back marked as the client’s, and the assistant is told to report them, never to act on them.

How do we see what it did, and disconnect it?

Anything an assistant records appears under the person’s name in OneHQ, with the same audit trail as if they had typed it, and authorising an assistant is itself recorded in the audit log. Each person can see their connections and their assistant’s recent activity in OneHQ.

To disconnect, open Settings → Integrations → My MCP connection, or My Profile, and disconnect the assistant there. Its next request is refused; nothing needs doing on the assistant’s side. A practice administrator can also disconnect a staff member’s assistant, and turning the practice’s switch off stops every connection at the firm on its next request. Access tokens last an hour and renew themselves, and OneHQ stores only a hash of each token, never the token itself.

Where does our data go, and is it used for training?

When your practice connects an assistant, the data that assistant asks for is sent to its provider, such as Anthropic or OpenAI, under the provider’s terms, and that provider is generally in the United States. Nothing is sent until one of your people authorises a connection, and what is sent is limited to what the tool the assistant called returns. OneHQ does not use client data to train, fine-tune or improve any AI or machine learning model. Sections 10 and 12.2 of our privacy policy say so in full.

Frequently asked