How much does Claude for Small Business cost?

Anthropic's announcement does not list a separate price for the package. It installs inside Claude Cowork, so the cost is the Claude plan your team uses; check which plan includes Cowork before you count on it.

Can Claude connect to software that isn't on Anthropic's list?

Yes, through a custom connector, technically an MCP server. Your system exposes a set of actions (find an order, update a record, draft a quote) that Claude can call, each one running with the permissions of the person signed in. We run our own CRM this way every day.

Is it safe to let Claude write to my business systems?

It can be, if the connector follows four rules: every call runs as the signed-in user rather than with a shared admin key, read and write are separate permissions, anything destructive shows a preview first, and access limits are enforced in the server rather than in the prompt.

Does Claude train on my business data?

Anthropic says it does not train on your data by default on its Team and Enterprise plans. Check which plan your team is on, and read the terms for the plan you actually pay for.

Claude for Small Business: What It Covers and What It Doesn't

Q: What is Claude for Small Business?

It is a package Anthropic launched on May 13, 2026: a toggle inside Claude Cowork that connects Claude to Intuit QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365, with 15 ready-made workflows and 15 skills across finance, operations, sales, marketing, HR and customer service. The owner starts every task and approves before anything is sent, posted or paid.

A chat panel wired to four small app tiles, with a new cable passing through a padlocked gate into a larger, older business system

Claude for Small Business is a package Anthropic launched on May 13, 2026. It's a toggle inside Claude Cowork that connects Claude to QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365, with 15 ready-made workflows like the monthly close, an invoice chaser and a lead triager. Anthropic's own summary is the right one: "Claude does the work; you approve before anything sends, posts, or pays." If your business runs on those tools, start there. If it also runs on a system that isn't on the list, that system needs a connector of its own, and the rest of this post is about how to build one safely.

What Claude for Small Business includes

From Anthropic's announcement:

Connectors to Intuit QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace and Microsoft 365.
15 agentic workflows and 15 skills across finance, operations, sales, marketing, HR and customer service. Named examples include payroll planning, the monthly close, a business pulse dashboard, an invoice chaser, a margin analyzer, a contract reviewer and a lead triager.
Owner approval on every action. You start each task and approve the plan, and your existing permissions in each tool carry over.
A free course and a tour. "AI Fluency for Small Business," built with PayPal, and an in-person Claude SMB Tour across US cities.

The announcement doesn't list a separate price. The package installs inside Claude Cowork, so what you pay is the Claude plan your team is on.

It's the cheapest experiment a small business can run this month. Switch it on, give it one real job, and see whether it saves an hour.

The tool that isn't on the list

Most of the small and mid-sized businesses we work with have one system that isn't on that list. It's the order spreadsheet with 90 tabs. The industry system the whole team logs into every morning. The old ERP nobody wants to touch. The app someone built five years ago that still runs the warehouse.

Claude can read your email and your QuickBooks. It can't see the system where your orders, quotes or jobs actually live, unless someone connects it. That connection is a custom connector, technically an MCP server. Without one, you're back to copying data from one window into another, only now one of the windows is a chat.

What it looks like when it works: our own CRM

In February 2026 we built our own CRM instead of customizing Salesforce. It's a Go and PostgreSQL application, and it has shipped more than 750 releases since. Two days after the first commit, we gave it a connector so Claude could read and write it directly.

We also built AI into the CRM itself: generated messages, a coaching tile that reads the pipeline, an AI recap on the dashboard. Both still exist. But the daily work moved to Claude, because that's where we already were. In-app AI is one more screen to open. The CRM in Claude is a question away.

Today the CRM exposes more than 160 tools to Claude. A normal morning:

"Who needs a follow-up today, and why?" The answer comes from live data, not from memory.
A call ends, I dictate two sentences, and the interaction is logged against the right contact and deal.
"Move the deal to proposal sent and draft the follow-up." Claude shows the change before it makes it.
"Plan my day." It builds the list from overdue follow-ups, meetings and open deals.

None of this is a chatbot. Each one removes a few minutes of clicking through screens to find something the system already knew.

Four rules that make it safe to let Claude write to your systems

The plumbing was the easy part. The months went into permissions.

1. It acts as you, not as an admin key. At first, connecting Claude meant pasting an API key. That works for whoever built it and for nobody else. Since August 2026 our CRM uses MCP OAuth: you add it to Claude as a custom connector, log in with your own account, approve a consent screen, and every call runs with your permissions. One page lists every assistant you've granted access to and revokes it in a click.

2. Read and write are separate permissions. You can give an assistant read access without letting it change anything. Any tool not explicitly marked read-only needs write access, so a forgotten label fails safe.

3. Anything destructive shows a preview first. Updates and deletes show what would change, including knock-on effects, before anything happens. Duplicate checks run before anything is created.

4. The rules live in the server, not in the prompt. Every call is limited to your own company's data on the server side. A model can be talked into anything. A database filter can't.

Anthropic's package makes the same choices: you start every task, you approve the plan, your existing permissions carry over. That's the right standard. Hold a custom connector to it too. If you want to try one first, here is how to add a custom connector in Claude.

How to start

List the three things your team retypes every week. Not "use AI." Specific jobs: entering orders from emails, chasing unpaid invoices, updating the pipeline after calls.
Check whether Claude for Small Business already covers them. If all three live in QuickBooks, HubSpot and Google Workspace, you may not need anyone like us.
If one lives in your own system, that's a connector job. Scope one workflow, on your real data, with a person approving every write, and measure it before adding the second.

Our published prices: scoping one workflow costs $2,300 to $4,700, and a first workflow in production starts at $17,500. Running it afterwards starts around $470 a month. Details here.

Questions to ask before you connect Claude to anything

Does it act as me, or with a shared admin key?
Can I give it read access without write access?
Does it show me a change before it makes it?
Where does my data go? On Team and Enterprise plans, Anthropic says it doesn't train on your data by default. Check which plan you're on.
Who maintains the connector when my system changes?
Can I see and revoke every assistant that has access, in one place?

Claude for Small Business: What It Does Once It's Connected to Your Own Tools

Anthropic's package connects Claude to QuickBooks, HubSpot, PayPal and the usual suite. Most businesses also run one system that isn't on that list, and that is usually where the work is. Here is how we connected ours, and the four rules that make it safe.