Is It Safe to Connect Claude to Gusto?

Connecting Claude to Gusto is safe enough if one named admin connects it, approves every tool call and keeps other connectors out of that chat. The listing undersells what you’re granting. It says “Query and analyze your Gusto data”. Its tools can also run payroll, link a bank account and create a company.

I haven’t tested this connector. Everything here comes from public pages I read on October 31, 2026.

The listing has 62 tools. Gusto’s docs describe 12 more that it doesn’t show, including IRS filings. The table of what 33 connectors can change compares it with your other apps.

What can Claude do in Gusto once it’s connected?

Mostly read. The 11 tools that don’t just read are the ones to worry about.

The other 51 read. The docs-only tools add file_ein, pay_ein_filing and file_scorp_election, which prepare IRS filings and create a checkout link to pay for one. A lot of this is a sign-up flow for new businesses. An existing 10-person company has no use for most of it.

The list keeps growing. Gusto’s changelog shows update_payroll arriving in July, the bank tool in August and EIN filing wired in during September. The listing also warns that “Anthropic does not control which tools developers make available”.

Can Claude run payroll in Gusto?

Yes, for some customers. The listing says “Eligible customers can also run payroll in Claude”, and Gusto’s changelog has a run_payroll for Claude live since February. You may find a page saying the connector is read-only, like this one. Gusto’s current docs list run_payroll and never use the words “read-only”.

Gusto’s docs give the tool one line: “Calculate and submit an existing unprocessed payroll.” So read whatever Claude shows you like a person who is about to move money. Neither tool list has anything that cancels a payroll, so a mistake gets fixed inside Gusto.

What employee data can Claude see in Gusto?

Salaries and home addresses, yes. Full Social Security numbers and employee bank details, I found no tool for.

Reports are the gap in that table. get_reports returns CSV files such as the payroll journal, and the published scopes include both company_reports:read and company_reports:read:excludes_sensitive_columns. So some report columns count as sensitive, and which version you granted matters. Pull a payroll journal in a demo company and read the columns before you trust the last two rows.

You pick data categories at sign-in, and this is the most useful control Gusto gives you. Each connection is “limited to only the data categories you explicitly grant during setup”. Blocking get_compensation won’t hide pay, because the salary example above runs through list_employees. In Gusto’s API, pay fields “are only returned when the compensations:read scope is included”, so leave pay out at sign-in unless the job needs it.

Who at the company can connect Gusto to Claude?

Only your top Gusto admins. Gusto’s docs say it “requires a Gusto account with global/full admin permissions”.

So the person connecting already holds the whole company’s pay. That keeps everyone else out, because Claude inherits each person’s permissions and Gusto stops a regular employee. The cost is scope. There’s no “payroll viewer” role to connect with instead.

On Team and Enterprise, a Claude owner enables the connector and sets its tools, then the Gusto admin signs in and picks categories. On Pro or Max, the admin does it alone. Nobody can set tool rules for the whole company or see that it happened.

Who sees payroll data once it’s in Claude?

The answer to “who are our highest paid employees?” is now a chat, and it lives outside Gusto.

Check that Primary Owner row against your org chart. If your Gusto admin is the office manager and your Primary Owner is the contractor who set up Claude, you just gave that contractor a route to every salary Claude writes into a chat.

Chats also stay. On company plans Anthropic says “we retain your chats and coding sessions in the product” until someone deletes them or, on Enterprise, a retention period does, so delete the chat when you have your answer. Anthropic says a deleted chat leaves its back end within 30 days, except where the law requires longer or its automated systems flag the content.

Employee pay and addresses are personal data. If a state law, handbook or contract says who may handle them, ask whoever owns that before connecting. This isn’t legal advice.

Gusto’s docs list “Export your tool usage logs” as one of your rights. Ask for them if you ever need to know what Claude called.

Is Gusto safe with AI?

Safe enough, if you follow Gusto’s own rules. Its security guidelines include four that shape the setup:

• Run it alone. “Do not connect other MCP servers to the same client session that is connected to Gusto”, because your data could be “inadvertently sent to another service.”
• Approve everything, reads included. It’s “critical that you don’t enable any automatic tool execution features”, and “You should always be prompted to confirm any tool call”.
• Watch web search. “Be especially cautious of AI features that perform external web searches”, since your data could “be sent to external search engines or third-party services.”
• Opt out of training yourself. Gusto’s no-training commitment “does not extend to whichever LLM client you choose”, and “opting out on your end is the only way”.

That last one depends on your plan. On Team and Enterprise, Anthropic says “we will not use your inputs or outputs from our commercial products” by default, unless you send feedback. On personal plans it’s a setting, and training data “does not include raw content from connectors”, though it may include what’s “directly copied into your conversation”. Claude’s written answer naming your salaries is your conversation. Turn the setting off.

The first rule is the awkward one. A company brain is useful because it joins your tools together, and Gusto is asking you not to do that here. I’d follow Gusto. Payroll questions get their own chat, with the other connectors and web search off.

How do I connect Gusto to Claude safely?

1. Try the demo first. Gusto’s guide says to use its demo company “before connecting your production company data”. Accounts come from dev.gusto.com, and the Claude server is https://mcp.api.gusto-demo.com/anthropic.
2. Set the tools in Claude, with an owner doing it for the organization. Users “can’t override it.”

3. Pick categories narrowly at sign-in. If the job is headcount and time off, leave pay out.
4. Run get_token_info and read which scopes are live. To change them, Gusto says to disconnect and reconnect.
5. Write down who connected, and disconnect in Customize, then Connectors when they leave. Gusto says tokens are “revoked on disconnect”.

What I couldn’t confirm

• Whether a payroll run always stops for a preview and your confirmation, including when an owner sets run_payroll to Always allow. Search summaries of Gusto’s Claude page say it’s designed to, but that page and Gusto’s help center returned 403 to me, so I never read the wording. The changelog calls the Claude tool “Widgetized” and doesn’t say what the widget shows.
• Whether a demo company can run a payroll through Claude. The demo guide only tests read questions.
• Whether any report or tool returns full SSNs or bank numbers. The API facts above describe Gusto’s REST API, and the connector could differ. That API’s employee record also has has_ssn and an undocumented hidden_ssn.
• Which sign-in categories map to which scopes, and whether you can decline writes separately.
• Whether Team is supported, and who counts as “eligible”. Gusto’s troubleshooting section names only “Pro/Max for Claude”, and Anthropic says web connectors are “available for all users”.

Should I connect Claude to Gusto?

Yes for read questions on a company plan, with pay left out unless you need it. For running payroll, I’d wait until you’ve done it in the demo and watched what it asks you. For the rules around it, use the one-page connector policy in the QuickBooks post. Its rule on regulated data covers payroll.

If you’d rather plan which of your tools to connect, and how, the setup planner asks six questions and gives you a plan without asking for an email. If you want a second look at a list you already have, bring your connected apps to a 30-minute call.