QuickBooks MCP Server Guide 2026: Options, Setup and Safety

A QuickBooks MCP server lets an AI app such as Claude, ChatGPT, Cursor or Grok read and work in your QuickBooks Online books. You can run a server yourself, use a general connector platform, or connect a hosted service such as DeepLedger in about five minutes. This guide compares the options honestly, walks through setup in each AI app, and shows what to ask once you are connected.

It is written for accountants, bookkeepers and finance teams, not developers. Where a step needs a technical detail, we say so and keep it short. If your books are in Xero, read the Xero MCP guide instead.

What is a QuickBooks MCP server, in plain words?

MCP stands for Model Context Protocol. It is a shared standard that lets an AI app plug into another system, the way a USB cable lets one device plug into another. The AI app is on one end, your books are on the other, and the MCP server sits in the middle.

The server offers the AI a list of tools. A tool is one specific job, such as "run a profit and loss report", "look up a vendor" or "record a bill". When you ask a question in plain English, the AI picks the tools it needs, the server runs them against QuickBooks, and the AI explains the result.

Three things follow from that:

• The server decides what the AI can do. If a server has no tool for bank reconciliation, the AI cannot reconcile, however well you phrase the request.
• Reading and writing are different tools. A report tool only reads. A bill tool changes your books. Check which kind a server offers before you connect it to a real company.
• MCP adds no accounting rules of its own. Duplicate checks, approvals and an audit record exist only if the server builds them.

MCP server or the QuickBooks API?

They are layers, not rivals. The QuickBooks Online API is Intuit's interface for programmers. Every QuickBooks MCP server uses it underneath.

The options compared

There are three realistic ways to put QuickBooks Online inside an AI app today. They differ in who runs the server, how many companies it can reach, and what happens around each write.

Details change often, especially for the platforms. Check each vendor's current page before you choose.

Self-hosting an open-source server

Several projects on GitHub wrap the QuickBooks API as an MCP server. To run one you register an app on Intuit's developer portal, run a sign-in step to get keys, and store those keys in a settings file. On your own computer it runs as a local program that a desktop AI app starts, and web and phone apps such as claude.ai or ChatGPT cannot reach it. Your team can instead run it on its own server so that web AI apps can reach it.

Your team then owns everything around it: storing Intuit keys safely, refreshing tokens, keeping companies apart, updating the code when Intuit changes its API, and keeping logs. That is a real job, not a one-time install.

Good for: firms with in-house developers and a specific need no hosted option meets.

Connector platforms

General connector platforms give AI apps hosted access to hundreds of business apps, QuickBooks among them. They handle the sign-in for you and give you one URL.

They are built for breadth, with QuickBooks as one app among many. Before you rely on one for bookkeeping, check what it offers for duplicate checks, approvals, keeping clients apart and an audit record, because these vary by platform. Pricing is per platform plan or per task.

Good for: automation builders who want QuickBooks as one of many apps in a workflow.

DeepLedger's hosted harness

DeepLedger is the accounting harness for AI: connect QuickBooks or Xero once, and every AI your firm uses works safely across all your clients, with your firm's rules, memory and review built in. A harness, in plain words, is the layer around the AI: who may do what in each company, what the AI should remember about each client, and where a person reviews and signs off.

In practice:

• One address works in every AI app: https://mcp.deepledger.ai/mcp.
• One sign-in reaches every company you can access. Tool results identify the company, so you can confirm it before relying on an answer.
• QuickBooks and workflow tools cover recording, 30 QuickBooks reports, the bank feed, month-end close, reconciliation, documents and saved report packs.
• It works when you ask. Nothing runs on its own: the AI acts only when you tell it to, and you can tell it to read the company's AI Brain and skills first.
• A person stays in charge. The AI is instructed to turn uncertain items into Tasks, the close ends with a person's sign-off, and every AI action lands in Audit history.

The honest trade-off: it is a paid service ($20 per connection per month after a 14-day free trial), and your data passes through DeepLedger as well as Intuit and your AI provider.

Good for: accounting firms, bookkeepers and in-house teams who run several companies and want the AI to do real bookkeeping with review around it.

Which one should you pick?

• You are a developer and want everything local: run a self-hosted server.
• QuickBooks is one step in a wider automation: a connector platform fits.
• You do the books for several companies and need review, memory and an audit record: use a hosted harness such as DeepLedger.

Set up DeepLedger's QuickBooks MCP in 3 steps

Before you start, you need a DeepLedger account (the 14-day trial needs no card), access to the QuickBooks Online company, and an AI app that can add a custom connector. The steps match the Connect AI agent page in the DeepLedger portal.

Step 1: Connect QuickBooks to DeepLedger

In the DeepLedger portal, open Companies, choose the company (or Add company), and open its Connections tab. Under Accounting software, choose Connect QuickBooks and sign in with Intuit. DeepLedger never sees your QuickBooks password; Intuit's own sign-in grants access.

Do this once per company. A firm with 40 clients connects 40 companies here, and every AI app then reaches all of them through the same address.

Step 2: Copy the MCP address

Open Connect AI agent in the portal's top bar, or copy the address from here:

https://mcp.deepledger.ai/mcp

Include the /mcp at the end. It is a public address, not a password, so it is safe to paste into any AI app.

Step 3: Add it to your AI app as a custom connector

In your AI app, add a custom connector (some apps call it a custom MCP server or plugin) with that address. When the app opens a sign-in window, sign in with your DeepLedger account and choose Allow. Then ask:

Which company am I in? List the companies I can access.

If the AI names your company, you are connected. If your app is not listed below, paste the address into it and ask, "How do I add this as a custom MCP connector?"

Adding DeepLedger to each AI app

Menu names change as apps update. These paths come from each vendor's own documentation, checked on November 6, 2026. If a menu has moved, the linked page has the current steps.

Claude (web, desktop and mobile)

1. Open Customize → Connectors, select + Add, then Add custom connector.
2. Name it DeepLedger QuickBooks and paste https://mcp.deepledger.ai/mcp. Select Continue.
3. Keep Sign in now. For the OAuth client, Register automatically works with DeepLedger. Select Add, then sign in to DeepLedger and allow access.
4. In a chat, turn it on from + → Connectors.

On Team and Enterprise plans, an owner adds it first under Organization settings → Connectors → Add → Custom → Web. Members then find it in Customize → Connectors and select Connect. Free accounts can add one custom connector. Source: Claude's custom connector guide. Our Claude walkthrough has screenshots.

ChatGPT

1. Go to Plugins, select the + button, then Add custom MCP server.
2. Name it DeepLedger QuickBooks. Under Connection, paste https://mcp.deepledger.ai/mcp, and choose OAuth for authentication.
3. Tick I understand and want to continue, then Create as a plugin. Install it, sign in to DeepLedger, and mention it with @ in a chat.

OpenAI currently gives full MCP, including write tools, to Business, Enterprise and Edu workspaces on the web; Pro accounts get read and fetch only. In Business workspaces, only admins and owners can create and publish these plugins. Source: OpenAI's custom MCP server guide. See also our ChatGPT walkthrough.

Cursor

Add this to ~/.cursor/mcp.json (all projects) or .cursor/mcp.json (one project), keeping any servers already there:

{
  "mcpServers": {
    "deepledger": {
      "url": "https://mcp.deepledger.ai/mcp"
    }
  }
}

Open the Customize page in Cursor's sidebar, turn DeepLedger on, and complete the sign-in when asked. Source: Cursor's MCP documentation. Our Cursor guide covers the API key option.

Grok

1. Open grok.com/connectors.
2. Select New Connector, then Custom.
3. Paste https://mcp.deepledger.ai/mcp, then sign in to DeepLedger when asked.

xAI says connectors are available to all Grok users. On Grok Business and Enterprise, a team admin adds it first in the xAI console under Grok Business → Connectors → Add Connector → Other. Sources: Grok connectors and connector management. Our Grok guide covers Grok Bot and the xAI API too.

For developers: custom agents and APIs

This part is for whoever builds your firm's own AI tools. Agents you build yourself use a DeepLedger API key instead of a sign-in window. Any team member can create one in the portal under Settings → API keys → Create key, for All my companies (the default) or One company. Keep it in an environment variable and pass it as the bearer token. The key acts as you, with your access: an all-companies key works in your active company, and a one-company key stays in that company.

• Anthropic Messages API: add the server under mcp_servers (type url, with authorization_token) and an mcp_toolset entry under tools, with the MCP connector beta header. See Anthropic's MCP connector docs.
• OpenAI Responses API: add a tool of type mcp with server_label, server_url and authorization, and keep require_approval on for writes. See OpenAI's MCP tool guide.
• xAI Responses API: the same mcp tool shape with server_url, server_label and authorization; xAI does not support require_approval. See xAI's remote MCP docs.

Framework examples for LangChain, CrewAI and others are in our custom agents guide.

What you can ask it to do

Start with something you can check against QuickBooks, such as a report, before asking for writes. Name the company, the dates and, for reports, cash or accrual. These prompts work in any of the apps above.

Categorise transactions

In Sandbox Bakery, list last month's uncategorised bank transactions. Suggest an account for each from how this vendor was recorded before. Record the ones you are sure of and open a task for anything you are not.

The AI is instructed to check the vendor's history in QuickBooks and the company's rules in AI Brain before suggesting an account. See fixing uncategorized transactions.

Reconcile a bank account

Prepare the October reconciliation for the main checking account. The statement ending balance is $48,210.55. Show what is cleared, what is missing, and any duplicates.

The reconciliation keeps its proof (statement balance, cleared items and the difference) for a person to sign off. See AI bank reconciliation.

Close the month

Start the October close for Acme Consulting. Run the checklist, flag anything unusual against September, and propose adjusting entries for my review.

The close runs as a checklist in Month-end close, and finishing it creates a sign-off task for a person. See the month-end close checklist.

Receivables and payables

Show A/R aging as of today. List invoices more than 30 days overdue with the customer, amount and last payment date. Draft a polite reminder for each, but do not send anything.

Here is a vendor invoice (attached). Check it is not already entered, then show me the bill you would record before saving it.

Sending an email to a customer is a separate action, never part of recording, and the AI is instructed to get your approval first. Also see finding duplicate bills and invoices.

Reports and analysis

Compare Q3 profit and loss with Q2 on an accrual basis. Explain the three biggest changes in plain English and show the transactions behind each.

Save a report called "Monthly review": last month's P&L against the month before, plus the balance sheet at month-end. Run it now.

When you rerun a saved report that uses relative dates, such as "last month", those dates update automatically; fixed date ranges stay as saved. For forecasting, see the 13-week cash flow guide.

Dashboards across companies

For each of my companies, show cash in the bank, revenue this month, open Tasks and whether the last close is signed off. Put it in one table.

Apps that draw charts, such as Claude and ChatGPT, can turn the same request into a dashboard you reuse.

Safety and review: how a person stays in charge

QuickBooks has no "draft" state for most transactions: a bill or expense is either recorded or not. So DeepLedger puts the review around the write instead. Some of it is guidance the AI follows, and some of it is enforced by DeepLedger whatever the AI does.

How DeepLedger instructs the AI to work:

• Read before writing. Before recording, look up the vendor or customer, their history, open bills and invoices, and anything with the same amount within about two weeks that could be a duplicate.
• Write only with a reason. Go ahead when you asked for it, when a person approved it in Tasks, or when the company's QuickBooks history clearly supports it. Otherwise open a Task with the reasoning instead of guessing an account.
• Stop on warning signs. A likely duplicate, an amount far above the vendor's usual, or a conflict (a bill already open when you asked for an expense) means asking you first.
• Correct without erasing. Void wrong entries where QuickBooks supports it, rather than deleting them.

AI models usually follow these instructions, but they are not a lock. That is why the controls below exist.

What DeepLedger enforces:

• Read-only members can ask anything but cannot change the books. Owners and admins can also make chosen areas, such as journal entries or voids, read-only for one member.
• The right company. Every write must name its company. If the company is missing or differs from your active company, DeepLedger refuses the write, nothing is recorded, and the AI has to ask you.
• No delete tool. Posted transactions cannot be deleted through DeepLedger. Voids need confirmation and keep the record.
• Customer emails go out only through a separate send action, never as a side effect of recording.
• Plan check. Accounting tools run only for companies on a trial or a plan.
• Two-step sign-in is available for every account; turn it on under Settings → Account.
• Audit history lists what the AI did in each company: which tool, when, by whom, and whether it worked.

Where people review:

• Tasks is the shared to-do list between your team and the AI (the list was called Inbox until November 2026). Questions, approvals and document requests land there.
• Month-end close ends with a sign-off by a person, and a reconciliation keeps its proof.

A prompt that says "don't change anything" is not the same as read-only access. For a client you only want to read, make the person a read-only member. More on this in AI write access and safety and read-only QuickBooks MCP access.

Working across many companies

Firms are where a hosted harness earns its keep. With DeepLedger:

• One sign-in, every company. Ask "List my companies" and "Switch to Acme Consulting". The switch applies to all your AI apps until you switch again.
• Every tool result names its company, so you can check whose books an answer came from before you use it.
• Rules stay with each company. Each company's AI Brain holds its vendor rules and notes, so "Uber is always Travel" for one client does not leak into another.
• One review queue. The portal's Companies page shows open Tasks and close status for every company at a glance.
• Pricing is per connection, so a small client's QuickBooks connection costs the same $20 a month as a large one's, with unlimited users.

If you run Intuit Enterprise Suite, the same sign-in works across its entities; see connecting Intuit Enterprise Suite to AI. For how company switching behaves across apps, see multiple companies, one AI sign-in.

Limits worth knowing

• QuickBooks Online only. No MCP server here reads a QuickBooks Desktop file. See Desktop vs Online for AI.
• Intuit's rate limits. Intuit allows about 500 API requests per minute per company, and 10 per second. DeepLedger adds its own limit of 100 tool calls per minute per company. A huge history pull is split into pages, so ask for a date range rather than "everything".
• Some features need one more approval. Projects and custom fields use Intuit's premium APIs. On QuickBooks plans that include them, the company's Connections tab offers Enable projects and custom fields, which takes one approval in QuickBooks.
• Your AI app has its own limits. Claude's Free plan allows one custom connector, and ChatGPT reserves write tools for Business, Enterprise and Edu workspaces. Long conversations can also hit the app's own message limits.
• Your AI provider sees what you ask. Report data in a conversation is handled under your AI provider's terms. Check your plan's data settings and get your firm's approval before using client data.
• The AI can still be wrong. Check the first few results against QuickBooks, and keep a person reviewing anything that matters.

Troubleshooting

Step-by-step fixes are in the QuickBooks MCP troubleshooting guide.

Frequently asked questions

What is a QuickBooks MCP server?

It is a service that gives an AI app a set of tools for working with a QuickBooks Online company, using the Model Context Protocol. The server decides which reports, records and write actions the AI can use. The protocol itself does not add any bookkeeping rules or review.

Do I need to run my own QuickBooks MCP server?

No. A hosted service such as DeepLedger runs the server for you. A server you run yourself works on your own computer or infrastructure, usually connects to one company per setup, and needs an Intuit developer app and OAuth keys that you set up yourself.

Is a QuickBooks MCP server the same as the QuickBooks API?

No. The QuickBooks Online API is Intuit's interface for programmers. An MCP server sits on top of that API and turns its operations into tools an AI app can find and use without custom code. Intuit's permissions and rate limits still apply underneath.

Which AI apps work with DeepLedger's QuickBooks MCP?

Any app that can add a remote MCP server, including Claude, ChatGPT, Cursor and Grok, plus agents you build with the Anthropic, OpenAI or xAI APIs. You add the same address, https://mcp.deepledger.ai/mcp, in each one and sign in with your DeepLedger account.

Can the AI change my books without me knowing?

Every tool call is recorded in Audit history, and every write returns what was recorded. DeepLedger instructs the AI to write only when you asked for it, a person approved it in Tasks, or the company's own history clearly supports it, and to open a task otherwise. Those are instructions the AI follows, not a lock, so for clients you only want to read, make the person a read-only member: read-only members cannot write at all.

Can one sign-in reach all my clients' companies?

Yes. One DeepLedger sign-in reaches every QuickBooks company you can access. You ask the AI to list or switch companies, and every tool result names the company it came from. Confirm the company before you rely on a report or approve a write.

Does it work with QuickBooks Desktop?

No. DeepLedger works with QuickBooks Online, including Intuit Enterprise Suite. QuickBooks Desktop needs a different kind of integration, such as Intuit's Web Connector.

How much does DeepLedger's QuickBooks MCP cost?

DeepLedger MCP is $20 per QuickBooks connection per month, with unlimited users, a 14-day free trial and no credit card needed. You use it with your own AI app subscription, such as Claude or ChatGPT.

Start with one company

Create a DeepLedger account, connect one QuickBooks company, add https://mcp.deepledger.ai/mcp to the AI app you already use, and ask for last month's profit and loss. Compare it with QuickBooks. When the numbers match, move on to the jobs above. Pricing is on the pricing page.