# Paychex MCP server for payroll admins

> A Paychex MCP server for payroll admins runs on one app credential, not a person. Role restrictions and the audit log decide who does what in Claude.

**TL;DR** Elaichi's Paychex connector signs in with OAuth client credentials, so Paychex sees one application, not the payroll admin who asked. Elaichi then carries the per-person control: a role restriction picks which of the 112 Paychex tools each admin can reach, and the audit log names the person behind every call that ran.

## Can a Paychex MCP server for payroll admins be limited to reads?

Yes. Connect Paychex in Elaichi with OAuth client credentials and add https://api.elaichi.ai/mcp in Claude, ChatGPT, Cursor or Microsoft Copilot. Then hold the payroll admin role to eight named read tools with a restriction on the role.

Payroll admins ask the same questions every cycle. Which pay periods are open. How many checks sit in this one. Was a new hire's I-9 status recorded. MCP (Model Context Protocol) is the standard way an AI assistant calls tools in other apps. The [Paychex connector](/connectors/paychex/) gives the assistant 112 Paychex tools to answer those, and the same connector can add a check, change a direct deposit or delete a pay rate. The design question is who gets which tools.

## Who is actually calling Paychex?

An application is, not a person. Paychex documents its API sign-in as a client credentials exchange: the application trades a client ID and secret for an access token, and the token authorizes the application to use the REST API (checked November 2026, [Paychex's authentication guide](https://developer.paychex.com/getting-started/authentication)). The OAuth specification describes that grant as the one for a client acting on its own behalf (checked November 2026, [RFC 6749](https://datatracker.ietf.org/doc/html/rfc6749)).

So when an admin asks Claude for the next pay period, the call that reaches Paychex carries the application's token, not their name. Everything else in this setup follows from that.

## Which Paychex access limits come from Paychex itself?

Paychex limits an application by company. Its companies endpoint returns the companies the application has access to, each with a hasPermission flag (checked November 2026, [Paychex's companies guide](https://developer.paychex.com/getting-started/companies)). A Flex admin approves a partner application for a company, then accepts the third-party terms and saves (checked November 2026, [Paychex's guide to linking client companies](https://developer.paychex.com/getting-started/integrate-with-paychex-clients)).

The sample token response lists scopes such as read:company_people, write:company_people and read:workers. Paychex's docs do not describe a per-person role for an API call, or a way to give one application read-only access (checked November 2026, same three pages). Ask Paychex support if you need that in writing.

Treat Paychex's control as a wall around companies. Anyone who can use the credential reaches everything inside the wall, so a second fence has to sit in front of it.

## Whose credential runs a Paychex call in Elaichi?

The connection's. In Elaichi, every call through a connection uses the credential that connection was made with, whoever is calling. With client credentials there is no personal sign-in behind it, so every payroll admin who can use the connection acts as the same Paychex application.

That leaves three choices for you:

1. Owner. Pick one payroll lead, not a departing contractor. If the owner leaves, transfer the connection to someone who stays first.
2. Who may run tools through it. Share it with the payroll admin team only.
3. Companies. Ask the Paychex admin to approve only the companies this team runs.

If you suspect misuse, disconnect the account. In Elaichi, revoking a share or disconnecting an account takes effect on the caller's next call, so every admin's next call stops.

## Which Paychex tools belong to which payroll role?

Sort the 112 tools by how far a mistake travels.

**Eight reads for every payroll admin.** list_all_paychex_companies, list_all_paychex_company_pay_periods, list_all_paychex_company_workers, paychex_company_workers_search, list_all_paychex_company_checks, list_all_paychex_worker_checks, list_all_paychex_company_pay_components and get_single_paychex_worker_i_9_status_by_id. They answer the daily questions and change nothing.

**Reads that show money details.** list_all_paychex_worker_direct_deposits returns routing and account numbers. list_all_paychex_compensation_pay_rates returns pay rates. Give these only to people who already see that data.

**Pay-run edits.** create_a_paychex_company_check and create_a_paychex_worker_check add checks. delete_a_paychex_worker_check_by_id removes an unprocessed one. paychex_checks_bulk_delete deletes the checks for a pay period and user. The check-component tools add, change or remove one earning, deduction or tax line. Keep these with a small payroll lead role.

**Money and tax records.** The direct-deposit create, update, bulk update and delete tools. The compensation pay-rate and worker pay-component tools. The federal and state tax tools. A wrong value here changes what lands in an account.

**Application management.** create_a_paychex_management_hook registers a webhook that sends events to an address you name. delete_a_paychex_management_remove_client_access_by_id unlinks client access. create_a_paychex_oauth_token requests a token. No chat role needs these.

## Allow rule or block rules for the payroll admin role?

Holding a role to an app's reads takes one of two shapes. One is an allow rule naming those reads, plus allow rules for every other app the role uses. The other is block rules naming every tool on the app that is not a read. With an allow rule, once a role carries any allow rule, every connector and tool none of its allow rules names is blocked for that role. With block rules, a layer with only block rules has no allowlist, so everything it does not block passes it.

The test is how many non-read tools you would have to list. Paychex has about 50 outside the 60-odd reads, 16 of them named delete or bulk_delete. That is too long to list without missing one, so the payroll admin role takes an allow rule naming the eight reads. Add whole-connector allow rules for the role's other apps, or it loses them when this rule saves.

In Elaichi, restriction targets are role or user only; the organization default is the absence of a rule, which means allow everything, so a second payroll lead role gets its own rule. A rule aimed at a member can only narrow what their role allows, and never replaces or loosens a role rule, which lets you narrow one admin without editing the role.

## What happens when an admin hits a limit, and what is logged?

An admin may ask for a Paychex tool the rule withholds. A restricted tool is left out of the tools Elaichi offers a client and cannot be called. Search still names it, flagged restricted, with no schema. The admin can file an access request, and a person decides it in the Elaichi console under Governance, then Access requests. A role change takes effect in about two minutes.

The audit log supplies what the shared credential cannot: the person. It writes one entry for each connected-tool call that reaches execution (a call refused earlier writes none). Each entry names the member, the tool, the connection and the AI client, and over MCP, a call from an MCP client is recorded under the person who signed in, with surface `mcp` and the OAuth client named. When a check appears in a pay period, you can ask who called create_a_paychex_worker_check and when. The trail records the one path argument that names the object, as the target id, and nothing else about the arguments, so pay data is not copied into the log. Elaichi keeps 90 days of history, which covers a monthly close. For longer, export to your own Datadog comes with the Black plan.

## How does a payroll admin set up the client?

Each admin adds https://api.elaichi.ai/mcp in their AI client and signs in through the browser. They approve their own grant. The address is the same for every organization, and the grant decides which organization and person the client reaches.

On that sign-in screen, "Use your connected apps" lets the client run every kind of Paychex tool, deletes included. No checkbox makes it read-only, so rely on the role restriction. For a second fence, choose "Only the ones I pick" and pick a toolbox of read tools.

Over MCP, Elaichi does not stop a write to ask first. A check added from chat runs as soon as the role allows it. If you want a confirm step before pay-run edits, turn it on in Claude or ChatGPT. In the client, connected tools are never listed one by one, however few there are, so the assistant searches for a Paychex tool and then runs it. Have the first admin try a read, then a write that should be refused.

## Does a one-person payroll shop need the Paychex connector governed?

Probably not the roles. If one payroll lead is the only person who will ever use Paychex from an AI client, connect it under that lead and keep the connection private. The Paychex application is still approved only for the companies that lead runs.

Governance pays off when several admins share the one application credential. Paychex cannot tell them apart, and the Elaichi audit log can. The [Paychex connector page](/connectors/paychex/) lists every tool, the [connector catalog](/connectors/) covers the other 900+ connectors, and [Gusto and Claude for an HR team](/blog/hr-team-claude-gusto-payroll/) shows a similar split for another payroll app.

## FAQ

### Does Paychex know which payroll admin made a call through Elaichi?

Not from the credential. Paychex documents its API sign-in as an OAuth client credentials exchange of a client ID and secret for a token (checked November 2026, https://developer.paychex.com/getting-started/authentication), which identifies an application. The person is recorded on Elaichi's side: the audit log names the member who signed in to the AI client.

### Can I make Paychex read-only for an AI client?

Yes, with a restriction, not a consent checkbox. In Elaichi, "Use your connected apps" runs every tier of a connected app's tools, and "Delete things" covers only Elaichi's own deletes. Write an allow rule on the role that names the Paychex read tools, or approve the connection with toolboxes that contain only read tools.

### Which companies can the Paychex connector see?

Only the companies the Paychex application has been granted access to. Paychex documents its companies endpoint as returning the list of all companies the application has access to (checked November 2026, https://developer.paychex.com/getting-started/companies). Company scope is therefore set in Paychex.

### How long until a restriction change reaches a payroll admin's AI client?

Within about two minutes. In Elaichi, a change to a role, a team or a restriction reaches every surface on that schedule. Removing or suspending a member makes their MCP clients fail on the next call. Disconnecting the Paychex account stops every call through that connection.

## Read next

- [Keep payroll data in HR with Gusto in Claude](/blog/hr-team-claude-gusto-payroll/) — Gusto in Claude runs on Gusto's own MCP server. One admin connects it in Elaichi, HR gets a Gusto-only toolbox, and roles decide who reaches which tool.
- [Video: Block a Risky Tool for a Whole Role, Once](/blog/block-a-risky-tool-for-a-whole-role/) — Block a risky tool for a whole role: one Elaichi restriction on Stripe's Create a coupon, set once for the Member role, holds in every AI app your team uses.
- [Audit Log Retention: AI Actions at 90 Days vs Seven Years](/blog/audit-log-retention-ai-actions/) — Elaichi keeps 90 days of audit log retention, AI actions included. A seven-year mandate needs a customer-owned destination, and only Datadog delivers today.
