Skip to main content

How to: Set up the XBert MCP

Connect Claude, ChatGPT, Claude Code or Microsoft Copilot to XBert.

Written by AI Support

What the XBert MCP is

MCP stands for Model Context Protocol, a standard, secure way to let an AI assistant work with another app on your behalf.

The XBert MCP connects your AI tool to XBert: your practice's work lists and tasks on one side, and your clients' live ledger data and alerts on the other.

In practice, you ask a question in plain English and your assistant answers it from your real data. Anything that would change the books is proposed for your review first, and nothing is applied behind your back.

You need one address, whichever assistant you use:

https://mcp-gateway.xbert.io/mcp

What you need

  1. An XBert account with at least one client on a paid plan.

  2. An AI Assistant that speaks MCP

    1. Claude on the web or Desktop, Claude Code, ChatGPT, or Microsoft Copilot via Copilot Studio.

    2. On a personal plan, you're ready to go.

    3. On a Team, Business or Enterprise plan, your organisation may have switched off the ability to add your own connectors.

      1. If Add custom connector is missing or greyed out, that's what has happened. Ask whoever administers your Claude or ChatGPT plan to allow custom connectors. This is your own IT or workspace admin, not your XBert admin.

  3. Your XBert login

    1. The assistant signs in as you, and sees only what you can see in XBert.

      1. If you cannot open and view a work item for a client in XBert, your assistant cannot read it either.

Which clients your assistant can see

Two things decide whether a client shows up via the MCP: whether you have access to it, and what plan it is on.

1. Your client access

Your assistant is limited to the clients on your Client Access Permissions list in XBert. An admin sets this per user, under Settings (the gear icon), then Users. It is the same list that controls which clients you see in the app.

If a client is not on your list, your assistant will tell you it cannot find that client. That is the permission working as intended, and rephrasing the question will not change it. An admin at your firm can add the client to your access in XBert.

There are four ways your access can be set:

  1. All Clients

  2. Selected Clients

  3. Xero Permissions + All Clients

  4. Xero Permissions + Selected Clients

Whichever one applies to you, your assistant sees exactly the clients that setting leaves you able to see in XBert, and nothing else.

2. The client's plan

Of the clients you can access, the MCP then follows the plan each one is on:

  • ✅ Workflow AI (Nano) - Your practice and workflow data for that client, with no ledger needed.

  • ✅ Financial AI - Everything above, plus that client's ledger data.

  • ❌ Manual - Not included. Your AI assistant will tell you the client is not on an AI-enabled plan.

  • ❌ Locked or lapsed trial - Not included until the client is on a paid plan.

So a client can be missing for either reason. If your assistant says it cannot find a client, that is access. If it says the client is not on an AI-enabled plan, that is the plan.

Set it up in Claude (web and Desktop)

Shortcut: if you are already signed in to Claude, this link opens Claude's Add custom connector box with the name and address already filled in. Then skip to step 5. On a Team or Enterprise plan an Owner has to do this — see the note at the end of this section.

  1. Open Claude at claude.ai or the Claude desktop app.

  2. Select Customize in Claude's left sidebar, then the Connectors tab. This moved out of Settings, so if you have been hunting through Settings for it, that is why it wasn't there. Some of Anthropic's own pages still call it Settings → Connectors.

  3. Choose Add custom connector.

  4. Name it XBert and paste the address: https://mcp-gateway.xbert.io/mcp

  5. Save it, then select Connect and sign in with your XBert account in the window that opens.

  6. Choose the business or Connect portal you want to work with.

  7. Back in the chat, XBert appears in your tools menu. Ask away.

The first time your assistant uses an XBert tool it asks permission. You can approve each time, or set it to always allow. These are the settings we recommend:

If Claude is not using XBert: the usual cause is that the connector is switched off for that conversation. Open the + menu in the chat, choose Connectors, and turn XBert on. Claude reaches XBert from Anthropic's own servers rather than from your computer, so there is no firewall or network setting on your side to change.

On a Team or Enterprise plan, an Owner goes first. Members cannot add a custom connector on those plans. An Owner adds XBert once for the whole firm under Organization settings → Connectors (labelled Admin settings in some places), and everyone else then finds XBert under Customize → Connectors and selects Connect. On Free, Pro and Max you add it yourself. Free allows one custom connector at a time.

Set it up in ChatGPT

In ChatGPT, custom MCP connectors sit behind developer mode, which is a web-only feature — it is not in the mobile apps. Where you find it depends on whether you are on a workspace or a personal plan, so follow whichever of the two sets of steps below matches your account.

Which plans can do this? OpenAI's own pages disagree. Its developer documentation lists Plus as eligible for developer mode, while its help centre lists only Business, Enterprise and Edu — so if you are on Plus and developer mode does not appear, that is why. On Pro, OpenAI documents read and fetch access only, which means XBert can answer questions but not take action. Write actions are a beta on the workspace plans. Free and Go are not listed by either page.

On a Business, Enterprise or Edu workspace

  1. An admin or owner opens Workspace settings and, under Permissions & Roles → Connected Data, turns on custom MCP connectors.

  2. Once you have access, open Settings → Apps → Advanced Settings and turn on Developer mode.

  3. Go to Apps → Create.

  4. Once the Create app screen appears, fill in the form:

    • Name: XBert

    • MCP server URL: https://mcp-gateway.xbert.io/mcp

    • Authentication: choose OAuth.

  5. Click Scan tools.

  6. Select Create. The connector appears as a draft with a Dev label — an admin or owner publishes it from Workspace settings → Apps → Drafts before the rest of the workspace can use it.

  7. Sign in with your XBert account when the browser window opens, and choose the business or Connect portal you want to work with.

  8. Start a new chat, turn on the XBert connector from the tools menu, and ask away.

On a personal Pro account

  1. In ChatGPT open Settings → Security and login and turn on Developer mode.

  2. Go to chatgpt.com/plugins and select the + button.

  3. Enter a name and description, and under Connection enter https://mcp-gateway.xbert.io/mcp

  4. Choose OAuth, sign in to XBert, then review the tools ChatGPT finds and create the connection.

OpenAI renames this area regularly — it has been called Connectors, Apps and now Plugins. If the wording does not match what you see, look for developer mode in your account settings, and for wherever custom MCP servers or apps are created. The address and the OAuth sign-in are the same either way.

Set it up in Claude Code

Try this first. If you have already added XBert on claude.ai, you may have nothing to do. Connectors added there load automatically in Claude Code when you are signed in with that same Claude account. Run /mcp and look for XBert before you add anything. This applies to a Claude subscription login, not an API key.

If it is not there, or you would rather register it locally, run:

claude mcp add --transport http --callback-port 6274 --scope user xbert https://mcp-gateway.xbert.io/mcp

  1. In your next session, run /mcp and complete the XBert sign-in when prompted.

  2. Ask in plain English.

--callback-port 6274 is not optional. XBert registers a fixed port to finish the sign-in, and Claude Code otherwise picks one at random — leave the flag out and the sign-in fails. --scope user makes XBert available from any folder rather than only the one you ran the command in.

If your Claude Code does not recognise --callback-port, update it first — the flag is only in recent versions.

Set it up in Microsoft Copilot

First, which Copilot do you mean? Microsoft uses the name for several products, and only one of them is a route to XBert.

  • Copilot Studio — Microsoft’s agent builder. This is the one. You add XBert as a tool on an agent, and it recognises our sign-in on its own.

  • Microsoft 365 Copilot (Copilot Chat in Teams, Word, Outlook) — indirectly. You cannot paste our address in here; Microsoft has no self-serve path for that. You reach it by building the Copilot Studio agent below and having an administrator approve it.

  • “Create agent” / Agent Builder inside Microsoft 365 Copilot — no. It takes knowledge sources only. Microsoft’s own advice is to use Copilot Studio when you need to reach an outside service.

  • Microsoft Copilot for consumers (copilot.microsoft.com) — no. Fixed list of connectors, no way to add your own.

  • GitHub Copilot in VS Code — no. A different product entirely.

Three things to sort out first

  • Generative orchestration must be on. In the agent that is Settings → Generative AI. New agents have it on already. Without it, MCP tools do not run at all.

  • Your IT administrator may need to act. MCP servers in Copilot Studio run over Power Platform connectors, so a data policy that blocks those also blocks XBert — quietly. Your admin can check in the Power Platform admin centre under Security → Data and privacy → Data policy.

  • Licences. Whoever builds the agent needs a Copilot Studio or Microsoft 365 Copilot licence. People who only use the finished agent do not need one. Worth knowing: the free trial can build and test an agent but cannot publish it. See Microsoft’s licensing guide.

Add XBert to a Copilot Studio agent

  1. Open your agent at copilotstudio.microsoft.com and go to its Tools page.

  2. Select Add a tool → New tool → Model Context Protocol.

  3. Fill in the form:

    • Server name: XBert

    • Server description: a sentence or two on what XBert does. Copilot reads this to decide when to call XBert, so it is worth writing properly.

    • Server URL: https://mcp-gateway.xbert.io/mcp

    • Authentication: OAuth 2.0, then Type: Dynamic discovery.

  4. Select Create, then Create a new connection, sign in with your XBert account, and select Add to agent.

You do not need a client ID, a client secret, or any URLs from XBert. That is what Dynamic discovery is for — XBert registers itself. Do not choose Manual: it asks for credentials we do not issue, and it will dead-end.

If your Copilot Studio signs you in at copilotstudio.com instead, you are on Microsoft’s newer experience and the path is Build → Tools → Add → Model Context Protocol (MCP).

Turn most of the tools off

XBert publishes far more tools than Copilot Studio will take. Microsoft caps an agent at 128 tools and recommends 25 to 30. Switch Allow all off and enable only the ones your practice actually needs, or the agent will not behave. Once Allow all is off, tools we add later arrive switched off too, so check back after an XBert release. If you are not sure which to pick, ask us and we will suggest a starting set.

Getting it in front of the practice

Publish the agent, then under Channels add Teams and Microsoft 365 Copilot. Expect one more step you cannot skip: the agent is not visible to your colleagues until an administrator approves it. Test it in the Test agent panel or in Teams while you wait — sign-in does not work on the Demo Website channel, so that one will look broken when it is not.

Microsoft’s own instructions: add an MCP server to an agent.

Any other MCP-aware assistant

  1. Find where your assistant manages connectors or MCP servers.

  2. Add a remote MCP server with the address https://mcp-gateway.xbert.io/mcp

  3. Complete the XBert sign-in when your browser opens, then start asking.

If your assistant asks for a transport type, choose HTTP (sometimes shown as "streamable HTTP"). If it asks for an OAuth client ID, leave it empty — XBert registers itself.

One catch worth knowing. An assistant that runs on your own machine finishes the sign-in on a port on your computer, and XBert accepts port 6274. Claude Code lets you set it (that is what --callback-port 6274 does above) and so does Goose (set GOOSE_OAUTH_CALLBACK_PORT to 6274). Assistants that use a fixed port of their own and give you no way to change it cannot finish signing in today — that currently includes VS Code and Cursor. Assistants that run in the cloud, like Claude on the web, ChatGPT and Copilot, are unaffected.

Your first five prompts

Start read-only. Get a feel for how it answers before you ask it to do anything.

  1. "What can you do with my XBert account?"

  2. "List the clients I look after."

  3. "Summarise this week's XBert alerts for [client]."

  4. "Who's over capacity next week, and what's running behind?"

  5. "What's on my worklist today?"

Name the client and the period. "Bayside Cafe, last quarter" beats "that cafe file" every time.

What you can ask it about

For any client on a paid plan, including Workflow AI (Nano):

  • Work lists, tasks, templates and schedules

  • Client notes and task conversations

  • Team capacity and upcoming deadlines

For clients on Financial AI, where a ledger is connected, everything above plus:

  • Invoices, quotes and receivables

  • Bills, purchase orders and payables

  • Bank activity, cash flow and reconciliation

  • Payroll and employees: pay runs, payslips, leave and retirement contributions (superannuation in Australia)

  • Reporting and the ledger: P&L, balance sheet, journals, budget vs actual, and tax and activity statement figures (BAS and IAS in Australia)

  • XBert alerts and exception review

  • Custom XBerts

  • Your own practice: jobs, timesheets and WIP, for firms running Xero Practice Manager

Each area is explained with example prompts, at intelligence.xbert.io/features.

Remember this isn't just one client at a time. Every client you have access to is there to query, across different ledgers, and XPM too. That's the power of XBert.

What it can and cannot do

Your assistant holds the permissions of the person signed in, and no more. It can see the clients on your Client Access Permissions list, read numbers and records when you ask, and propose actions for you to approve.

  1. It cannot see a client that is not on your client access list.

  2. It never learns your XBert password, because you sign in with XBert itself.

  3. It cannot change the books without showing you first: write actions are validated, checked for duplicates and proposed for your review.

Every action is recorded with an audit trail.

Managing and removing access

  • Turn it off in one place.

    • Remove or disable the XBert connector in your assistant's settings and the connection is gone. Reconnect any time by signing in again.

  • Access follows your client access.

    • Change someone's Client Access Permissions in XBert and their assistant's reach changes with it. Nothing is granted beyond it.

  • More than one business?

    • Ask your assistant to list your available businesses, then switch between them.

  • Sign-in uses OAuth, so your credentials stay with XBert and the assistant never sees your password.

Getting good results

  1. Start with questions, not changes

    1. Spend your first week asking: balances, alerts, work lists. Move on to drafting tasks and write-backs once the answers have earned your trust.

  2. Be specific

    1. Client, period, and what good looks like. Specific asks get specific answers.

  3. One client before twenty

    1. Prove a routine on a single file, then let the assistant repeat it across the portfolio.

  4. Review anything client-facing yourself

    1. Always. The assistant drafts; you decide what leaves the practice.

Need help?

If you have any questions, our support team is here to help!

Reach out via the chat widget on the bottom left of your XBert screen (Help), or email us at support@xbert.io

Related Articles
Did this answer your question?