Your HR data, now in ChatGPT, Claude and other AI assistants: the PeopleForce MCP server for managers, HR and recruiters
In March your plant ships its first batch to a distributor in Germany, and before that you need to pass CE certification. You don't just need people who are free. You need people who have already been through a European certification and negotiated with international partners. Hiring from outside takes three months and a serious budget. And you suspect those people already work in other departments — nobody just knows about their experience. You ask your AI assistant: “Who in the company speaks German, has the ‘CE certification’ skill and the top scores in the ‘Negotiating with international partners’ competency over the last two cycles, has been with us for over a year — and which of them isn't on leave in March?” A minute later you have a shortlist with names, departments, scores and time-off dates. Before, a list like that took days of asking around, and talent in other departments stayed invisible. This article is for people who already work with ChatGPT, Claude or another AI assistant and want it to know the team as well as your HR system does.
AI assistants reason well, but an answer is only as accurate as its context. The context about people — who reports to whom, who's on leave, how the last review went — lives in your HR system. PeopleForce is an HR platform that brings together employee records, time off, performance reviews, hiring and learning for more than 1,800 companies. MCP (Model Context Protocol) is an open standard that AI assistants use to connect to business systems. The PeopleForce MCP server gives your assistant access to your data — exactly within your permissions.
First, the most important part: the assistant acts on your behalf and sees exactly what you see. Anonymous feedback stays anonymous, private notes stay private, and decisions are always made by a person. More on this in “The assistant sees exactly what you see.”
If you're already a customer and want to connect right away, skip to “Connect in three steps.”
The interesting part starts where the answer doesn't fit on one screen. The assistant pulls data from several places, puts it together and shows you what usually goes unnoticed.
If your assistant is also connected to your calendar, Jira or Slack, it combines PeopleForce data with them in one answer:
We believe people data is the most sensitive data a company has. That's why the PeopleForce MCP server has no “special” access: the assistant acts on your behalf and within the same permissions you have in the web app or the mobile app.
Everyone signs in with their own account. The connection uses the OAuth 2.1 standard: the assistant never sees your password. It gets a short-lived key that you can revoke at any time.
Permissions apply to every request — by person and by field. If you can't see someone's salary or date of birth, neither can the assistant. This applies even to calculations: without access to dates of birth, the assistant can't group people by birth month.
A person decides what happens. The assistant creates a time-off request only after you confirm it. Review answers are saved only as a draft: only a person can submit, finalize or release a review, in PeopleForce.
Anonymous stays anonymous, private stays private. Anonymous feedback from colleagues arrives without any information about the author. A private 1:1 note is visible only to its author — neither HR nor an admin can open it. Colleagues' answers that haven't been released yet stay closed.
Colleagues' text is data, not commands. The assistant receives comments and posts written by people with an instruction not to follow anything they say. This protects against attempts to steer the AI through text.
You're in control. An admin turns the integration on and off for the whole company. Everyone can see and revoke their own connected assistants under “Preferences” → “Security” → “Connections,” and admins see everyone's in the “MCP connections” list. When someone leaves the company, their connections are revoked automatically. Your company's IP restrictions apply to MCP too: if sign-in is only allowed from the corporate network, the assistant can only connect from there. A per-person request limit stops an assistant that gets stuck in a loop.
Tested from the outside. The MCP server is in scope of our annual independent penetration test.
Some data we deliberately left outside MCP — even for people who can see it in PeopleForce.
Private messages in PeopleForce chats: these are someone else's private words.
Your conversations with AI Companion, the assistant built into PeopleForce.
Candidate contacts, CVs, interview notes and scorecards. These people don't work for you yet. The assistant answers about the pipeline, stages and timing — with names, but without contact details and without the content of CVs or scorecards.
Private agenda items and private action items from 1:1 meetings.
And one more thing we'd rather say up front: the data the assistant receives in its answers is processed by your AI provider under its own terms. For work with people data, choose your assistant's business plan, where data isn't used to train models.
Step-by-step instructions with screenshots are in the help center: “Connect an AI assistant with MCP”.
For automations where nobody sits in front of the screen, there's a second option: a service account with its own role and API key. You set its permissions the same way as for any other role.
In none of the questions above did you have to explain to the assistant who reports to whom, which department someone is in or which leave policy applies to them. Core HR — the PeopleForce foundation module — already knows that. Time off, performance reviews, hiring and reports run on the same foundation, and so does AI Companion. That's why your assistant's answer matches what you see in PeopleForce: it's the same data and the same permissions.
Yes. It's built into PeopleForce and runs at your workspace address with /mcp at the end. It's built and maintained by the PeopleForce team — not a third-party wrapper around the API.
Any that support MCP and OAuth sign-in: ChatGPT, Claude (including Claude Code) and others. Each user signs in themselves — no tokens to paste by hand. For automations without a person involved, use a service account.
AI Companion is the assistant inside PeopleForce. MCP brings PeopleForce data into the assistant your team already uses and lets you combine it with your calendar, Jira or Slack in one answer. Permissions are the same in both cases.
Only two things: create a time-off request after you confirm it, and save a review draft. Everything else is read-only.
The Professional plan, at no extra cost. MCP is enabled per account: if it isn't in your settings, contact your PeopleForce account manager, and then your admin turns it on.
No. An admin turns on the integration, and each user adds the URL to their assistant and signs in with their own account.
Any: English, Ukrainian, Polish, Spanish. The assistant answers in the same language.
The AI chatbot built into PeopleForce answers policy, people, leave and chart questions from your own data, within each person's permissions. What it does today and what's next.
Compensation components bring base compensation, bonuses, commissions and adjustments into one directory. This guide explains every concept, shows the setup step by step and what changes after the upgrade.
Not every employee works behind a desk. The redesigned PeopleForce mobile app brings everyday HR to the device frontline employees already carry, with sign-in by mobile number and no corporate mailbox required.