--- title: "ChatGPT and Xero: what connects and what to export" url: https://usingaias.com/ai-for/xero/ summary: "Xero has a plugin in ChatGPT (live since 15 September 2026), a read-only connector in Claude, and a developer MCP server with write access. Otherwise, export a report and strip the identifiers." published: 2026-10-01 updated: 2026-10-01 author: "Jack Stovell" publisher: "Adapt Progress Evolve Limited" language: en-GB --- # ChatGPT and Xero: what connects and what to export Xero has a plugin in ChatGPT (live since 15 September 2026), a read-only connector in Claude, and a developer MCP server with write access. Otherwise, export a report and strip the identifiers. ## At a glance - Difficulty: Beginner (The connectors and exports suit any Xero user. The MCP server is a developer project and is flagged as one.) - Time: About 15 minutes to connect, or 30 minutes for an export cleaned and ready (our estimate, not a measured figure) - Tools: Xero, ChatGPT or Claude, Microsoft Excel or Google Sheets - You need: Your organisation's AI policy, or a decision from whoever approves software and data use; The client's agreement, if it's a client's Xero organisation; The administrator, standard + reports or viewer role in Xero, to export reports - Role hub: [AI for accountants and finance managers](https://usingaias.com/ai-for/accountants/) ## What you'll end up with You'll know which route fits a given question: a live connection or an export. And when it's an export, you'll have a clean file with the identifiers taken out, ready for ChatGPT or Claude without handing over more than the question needs. The map below is correct as of 1 October 2026. This is moving quickly, so check each option's current status on Xero's own pages before you rely on it. ## Before you start Read your organisation's AI policy first. If there isn't one, ask whoever signs off software and data decisions. Don't assume silence means yes. Then check what role you hold in Xero. To export reports you need the administrator, standard + reports or viewer role. And if you're working in a client's organisation, connecting it to anything is the client's call as well as yours. It helps to know the names. JAX ("Just Ask Xero") is Xero's own AI inside Xero; among other things it does automatic bank reconciliation. The ChatGPT plugin and the Claude connector bring Xero data out to a general assistant. The MCP server is something else again, for developers. ## Steps ### 1. Decide: live data, or an export? Ask what the question actually needs. "Who owes us money right now?" wants live data. "Why did the margin move between these two periods?" is happy with a snapshot, and a snapshot is easier to control. A rule of thumb: if the answer changes by the hour, connect. If it's a point-in-time analysis you'll check and file, export. An export also leaves you with something you can keep in the working papers, which a chat doesn't. **Checkpoint:** you can write down, in one sentence, whether this question needs live data or a snapshot. ### 2. If you use Claude: connect the official Xero connector The Xero connector in Claude is read-only. Claude can view and retrieve your data, but it can't make changes to your Xero organisation. It covers revenue and profit, receivables and payables, financial and cash position, and invoices, and its answers link back to Xero so you can click through to the source. It's available on any Xero business pricing plan, it's free to add, and it needs a Claude account (free or paid). It connects to one Xero organisation at a time: to ask about another, you disconnect and reconnect. To set it up: open the connector directory in Claude, find and open Xero, click Connect, sign in to Xero if asked, choose the organisation and click Allow. Xero says data accessed through the connector isn't used to train Claude's model. Claude's Small Business plugin adds ready-made skills on top, such as a business snapshot, a cash flow forecast, a report builder and invoice chasing. That needs a Claude Pro, Max, Team or Enterprise plan. **Checkpoint:** Claude shows Xero as connected to the right organisation, and a simple question about who owes you money comes back with a link to Xero. ### 3. If you use ChatGPT: check the Xero plugin before you connect Xero announced on 15 September 2026 that its plugin in OpenAI's ChatGPT is live. It's powered by JAX, and Xero says it works across ChatGPT Chat, Work and Codex, using real-time Xero data across profit and loss, balance sheet, invoices and bills. Accountants and bookkeepers work in one client organisation at a time. Check two things before connecting. First, that the plugin is available on your own ChatGPT account; go by what Xero and OpenAI's current pages say, not by the announcement. Second, what it can reach once connected. If you need data outside the areas Xero lists, don't assume it's covered. **Checkpoint:** you've confirmed it's available to you, you know which Xero data it can reach, and you're allowed to connect this organisation. ### 4. Treat the MCP server as a developer project, not a shortcut You'll see the Xero MCP server mentioned, and it can sound like the grown-up version of the connector. It isn't. It's an open-source developer tool from XeroAPI on GitHub, under the MIT licence, that bridges the Model Context Protocol to Xero's API. It needs Node.js and a Xero developer account with API credentials, set up as a custom connection or a bearer token. And here's the important difference: unlike the Claude connector, it includes write commands. It can create invoices, bank transactions, manual journals and payments. That's write access to a live ledger, driven by an AI. Xero itself describes the connector as the read-only way to get answers and the MCP server as a developer tool for custom workflows. Take that at face value. If you do go down this road, run it as a proper project: someone technical, Xero's demo company first (the project's own instructions recommend starting there), and a named owner for the credentials. Those credentials are keys to the ledger, so they don't go in shared files. **Checkpoint:** either the connector or an export covers your need, or you've named an owner, you're testing in the demo company, and the credentials are stored securely. ### 5. Export the right report When a live connection isn't the right fit, export. In Xero, go to the Reporting menu, select All reports, open the report, click Export and choose PDF, Microsoft Excel or Google Sheets. A few reports earn their place: - **Trial Balance** shows general ledger account balances at your chosen date (accounts with nil balances don't show). You can add monthly columns and comparison periods, which helps when you're looking at movements. - **Account Transactions** gives you the transaction detail behind a balance. - **Aged Receivables Summary** and **Aged Payables Summary** show who owes what, who you owe, and how long it has been outstanding. - **Uncoded statement lines** (from Bank accounts, not Reports) exports unreconciled bank statement lines to CSV, for reconciliation questions. Pick a spreadsheet format where you can. You'll need to clean the file in the next step, and that's far easier in a spreadsheet than in a PDF. **Checkpoint:** you have the smallest set of reports the question needs, in a spreadsheet, saved where your policy allows. ### 6. Strip identifiers before you upload an export Before anything goes into ChatGPT or Claude, take out what identifies people and businesses: contact names, invoice and bank references, bank account details, and anything in a description column that names a person. You don't need to do it by hand. Ask the AI for the formulas, describing the layout only, and apply them yourself. ```text I have a spreadsheet exported from my accounting software. Column [A] holds contact names, column [B] holds invoice or bank references, and column [C] holds free-text descriptions. I want to replace each distinct contact name with a code like "Contact 001", using the same code every time the same name appears, and blank out the references. Give me Excel formulas I can paste in, tell me where the lookup table should go, and tell me what to check afterwards so no names are left in column [C]. ``` Then check the result yourself. Search the cleaned file for a few real names you know are in it. If one turns up, the clean isn't finished. Keep the lookup table (codes to names) separate from the upload; it's what lets you translate the answers back. **Checkpoint:** you've searched the cleaned file for known names and references, found none, and stored the lookup table away from the upload. ### 7. Ask for analysis, then check every figure in Xero Now ask for the analysis, and be specific about what you want and what the AI must not do. ```text Here is a cleaned export of [REPORT NAME] for [PERIOD], with contacts replaced by codes. Identify the [THREE] largest movements between [PERIOD ONE] and [PERIOD TWO], list any balances that look unusual, and suggest [FIVE] questions I should ask about them. Do not state any figure that isn't in the data I've given you. For every figure you quote, tell me which row and column it came from. ``` The numbers come from Xero, not from the AI. So go back into Xero and trace every figure you'll use to its source. If a total doesn't match, don't ask the AI to try again. Go to the ledger, find out why, and fix your understanding or the export. **Checkpoint:** every figure you'll use has been traced back to Xero, and anything that didn't match has been explained. ## Common mistakes - **Assuming a third-party connector is official.** Plenty of tools connect Xero to ChatGPT and Claude. If it isn't on Xero's own pages, treat it as a third-party tool and vet it like one. - **Connecting a client's organisation without asking.** Your access to their Xero isn't their agreement to send their data through an AI tool. Ask first, and keep a note that you did. - **Trusting a figure without clicking through.** The links back to Xero are there so you'll use them. - **Giving write access you don't need.** Read-only answers most questions. The MCP server's write commands change real records. - **Leaving MCP credentials in shared files.** It's the quickest way to turn a useful experiment into an incident. ## When to keep AI out of it Anything your policy rules out. Anything a client hasn't agreed to. And anything where you couldn't explain afterwards what was shared and why. If an export, a cleaned file and an AI answer all feed into your working papers, you need to be able to show where each number came from. If you can't, leave the AI out of that piece of work. ## Review checklist - [ ] The route (live connection or export) fits the question - [ ] Any connection used is an official one, checked on Xero's own pages - [ ] The right Xero organisation is connected - [ ] The client agreed, if it's a client's organisation - [ ] No write access granted that the task doesn't need - [ ] Exports in a spreadsheet format, saved where your policy allows - [ ] Contact names, references and bank details stripped before any upload - [ ] Searched the cleaned file for known names and found none - [ ] Every figure you'll use traced back to Xero - [ ] MCP credentials, if any, held by a named owner and not in shared files ## Before you paste anything: confidentiality and UK GDPR Before you paste anything, ask two questions: is there personal data in it, and is it confidential to my organisation or a client? If the answer to either is yes, use a tool your organisation has approved, under a business contract, and send only what the task needs. **For this task:** Don't upload Xero exports containing contact names, invoice or bank references, bank account details or staff payroll data to a consumer AI tool. Strip them first, and keep the code-to-name lookup out of the upload. Never paste these into a personal (consumer) AI account: - Payroll reports, salaries by name, bank details or National Insurance numbers - Named employee records: health, absence, grievance or disciplinary details - Customer or supplier ledgers with names attached - Unpublished results, forecasts or board papers - Anything a client has given you - Passwords, API keys or bank logins (in any tool, ever) UK GDPR, in four lines: - **Send the minimum.** UK GDPR's data minimisation principle says personal data must be adequate, relevant and limited to what is necessary for the purpose. For most tasks on this site, the personal data the AI needs is none. ([ICO: The data minimisation principle](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/data-protection-principles/a-guide-to-the-data-protection-principles/data-minimisation/)) - **Know who is controller and who is processor.** Under a business contract, the AI provider usually acts as your processor. On a personal consumer account, you are agreeing to the provider's own consumer terms instead. ([ICO: Controllers and processors](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/controllers-and-processors/)) - **Check where the data goes.** Many AI services process data outside the UK. Restricted transfers need safeguards, which a business agreement usually addresses and a personal sign-up does not. ([ICO: International transfers](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/international-transfers/)) - **The ICO has AI-specific guidance.** It covers accountability, transparency, accuracy and security when organisations use AI with personal data. ([ICO: Guidance on AI and data protection](https://ico.org.uk/for-organisations/uk-gdpr-guidance-and-resources/artificial-intelligence/guidance-on-ai-and-data-protection/)) Consumer plans vs business tiers (checked 2026-10-01; providers change their terms, so read the live page and your contract): - **OpenAI (ChatGPT).** Personal plans: conversations can be used to train models unless you turn off "Improve the model for everyone" in Settings > Data controls. ChatGPT Business, Enterprise, Edu and the API: not used to improve models by default. ([OpenAI Help Centre: How your data is used to improve model performance](https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance); [OpenAI: Enterprise privacy at OpenAI](https://openai.com/enterprise-privacy/)) - **Anthropic (Claude).** Free, Pro and Max: chats are used to train models only when the model-improvement setting is on. With it on, data is kept for up to five years; with it off, the standard is 30 days. Claude for Work and the API: inputs and outputs are not used to train models by default. ([Anthropic: Updates to Consumer Terms and Privacy Policy](https://www.anthropic.com/news/updates-to-our-consumer-terms); [Anthropic Privacy Center: Is my data used for model training? (consumer)](https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training); [Anthropic Privacy Center: Is my data used for model training? (commercial)](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training)) - **Microsoft (Copilot).** Personal Microsoft accounts are covered by Microsoft's consumer terms, not your organisation's. Copilot and Copilot Chat used through an organisation: covered by Microsoft's Data Protection Addendum with Microsoft as processor; your data is not used to train foundation models. ([Microsoft Learn: Enterprise data protection in Microsoft Copilot and Copilot Chat](https://learn.microsoft.com/en-us/microsoft-365/copilot/enterprise-data-protection); [Microsoft Learn: Data, privacy and security for Microsoft Copilot](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy)) > General information for UK readers. Not financial, legal, tax or HR advice. Example companies and figures are fictional unless a source says otherwise. Follow your organisation's policies and your professional body's rules. ## From the same studio Ours: made by the same studio that runs this site. - [Can You Use AI](https://canyouuseai.com/questions/do-my-bookkeeping/): Whether you can use AI to do your bookkeeping, and what HMRC and Making Tax Digital still expect of you when you do. ## Related - [AI for accountants and finance managers](https://usingaias.com/ai-for/accountants/) - [AI for bank reconciliation: the finance-team process](https://usingaias.com/ai-for/bank-reconciliation/) - [AI cash-flow forecasting: a weekly forecast you can defend](https://usingaias.com/ai-for/cash-flow-forecasting/) - [P&L with AI: build one from a trial balance](https://usingaias.com/ai-for/p-and-l/) ## Sources - [Xero Blog: Xero in Claude for Small Business and ChatGPT (15 September 2026)](https://blog.xero.com/product-updates/claude-for-small-business-chatgpt-ai-integrations-launch/) - [Xero Central: Connect Xero using the Xero connector in Claude](https://central.xero.com/s/article/Connect-Xero-using-the-Xero-connector-in-Claude) - [Xero UK: Claude, the Xero connector vs the Xero MCP server](https://www.xero.com/uk/ai-in-accounting/claude/) - [Claude: Xero connector (read-only, tool list)](https://claude.com/connectors/xero) - [Xero media release: Xero and Anthropic partnership (27 March 2026)](https://www.xero.com/us/media-releases/xero-and-anthropic-collaborate/) - [GitHub: XeroAPI/xero-mcp-server](https://github.com/XeroAPI/xero-mcp-server) - [Xero UK: JAX (Just Ask Xero)](https://www.xero.com/uk/jax/) - [Xero Central: Export and print a report](https://central.xero.com/0/article/Export-or-print-a-report) - [Xero Central: Trial Balance report](https://central.xero.com/0/article/Trial-Balance-report-new-version) - [Xero Central: Account Transactions report](https://central.xero.com/0/article/Account-Transactions-report-New) - [Xero Central: Aged Receivables Summary report](https://central.xero.com/0/article/Aged-Receivables-Summary-report-New) - [Xero Central: Export a list of uncoded bank statement lines](https://central.xero.com/0/article/Export-a-list-of-uncoded-bank-statement-lines) ## Questions people ask ### Does Xero work with ChatGPT? Yes. Xero announced on 15 September 2026 that its plugin in ChatGPT is live, powered by JAX and usable across ChatGPT Chat, Work and Codex, with Xero data across profit and loss, balance sheet, invoices and bills. Check it's available on your own account before relying on it. ### Is the Xero connector in Claude read-only? Yes. Claude can view and retrieve Xero data but can't make changes to your Xero organisation. It connects to one organisation at a time, is free to add on any Xero business plan, and needs a free or paid Claude account. ### What is the Xero MCP server? An open-source developer tool from XeroAPI on GitHub that bridges the Model Context Protocol to Xero's API. It needs Node.js and a Xero developer account, and unlike the Claude connector it includes write commands such as creating invoices, bank transactions, manual journals and payments. ### Which Xero reports should I export for AI analysis? Trial Balance for account balances at a date, Account Transactions for the detail behind a balance, Aged Receivables and Aged Payables for who owes what, and uncoded bank statement lines (CSV) for reconciliation questions. Export from Reporting, All reports, Export. ### Is my Xero data used to train AI? Xero says data accessed through its Claude connector isn't used to train Claude's model, and that data shared between Xero and Claude is used for that session only. For any other tool, check its own terms and your plan's settings.