MCP for HR: How to Connect People Data to AI¶

HR runs on questions that cross systems. How many people joined this quarter, how many left, and what did the recruiting spend look like against those hires? Which roles have been open the longest? Who has not completed onboarding training? The answers live in the HRIS, the applicant tracker, the finance system, and email. The Model Context Protocol (MCP) lets ChatGPT, Claude, and Perplexity pull those answers together in plain English, read-only, with the source named.

Where HR time goes¶

Headcount reporting, recruiting updates, onboarding checklists, and audit requests all take the same shape: pull from several systems, reconcile by hand, format for someone else.

The leadership update needs headcount and attrition by department
The hiring manager wants the pipeline for one role across two systems
Finance asks what recruiting actually cost against the plan
A compliance request needs onboarding records for a group of hires

Each one is manual assembly. MCP makes it a question.

Headcount, attrition, and org questions¶

Connect the employee system of record and the finance system:

"What is headcount by department at the end of each month this year, and where did it change?"
"What is our voluntary attrition rate this quarter, broken down by tenure band?"
"How many open roles are backfills versus net new hires?"
"Which departments grew fastest over the last four quarters?"

Because the answer is assembled from the live systems, the leadership number and the manager number come from the same source.

Recruiting pipeline visibility¶

If your applicant tracking data reaches a connected system, the pipeline question stops being a weekly export:

"How many candidates are at each stage for the engineering roles, and how long have they been there?"
"Which requisitions have had no candidate activity in ten days?"
"What is our time to hire by role and by source?"
"Which sources produced the candidates who reached final stage?"

Onboarding and offboarding¶

These are checklist problems that span systems, which is exactly where handoffs break.

"Which new hires from the last 60 days are missing an account, a device, or a completed form?"
"For each person who left this quarter, confirm that their accounts were deactivated and their device was returned."
"Which onboarding tasks are past due across all starters?"

The offboarding question in particular is a join between HR records, an asset register, and account data. That join is what MCP is for.

Compensation and people cost¶

Compensation data is sensitive, so the control that matters is access, not convenience. CorpusIQ does not hold your data and nobody at CorpusIQ can see your accounts.

"What is total people cost by department this quarter, including employer contributions?"
"How does average tenure compare across teams?"
"What is the distribution of time since last role change?"

Access to compensation fields follows the system you connect and the permissions you authorize. Everyone sees exactly what their own credentials allow.

Engagement and retention signals¶

Where attendance, booking, and communication data are connected, retention signals become visible earlier:

"Which teams have had a drop in booking activity this month?"
"Which managers have the highest volume of one-to-one cancellations?"
"Which newly promoted people have taken on significantly more direct reports?"

How CorpusIQ fits¶

CorpusIQ is the connector layer between the people systems you run and the AI assistant you already use. Every retrieval is read-only and cited.

Employee and operations records. Connect Odoo, which carries HR and operations records alongside CRM and finance.
Directory and documents. Connect Google Workspace or Microsoft 365 for directory, mail, and policy documents.
Hiring pipeline. Connect HubSpot or another CRM-shaped system that holds candidate records.
Scheduling. Connect Calendly for interview and one-to-one booking data.
Communication. Connect Slack for engagement and announcement context.
Structured HR data. Connect Airtable or a supported database for registers that live outside a dedicated HRIS.
People cost. Connect QuickBooks for payroll-adjacent and contractor spend.

Frequently Asked Questions¶

Can the AI see salary information?

Only what the credentials you authorize can already see. You connect each system yourself through its own OAuth flow, and the permissions are the ones you grant. Direct MCP requests fetch live; raw files and full response payloads are not warehoused.

Can the AI change an employee record?

No. The retrieval tools documented here are marked read-only. The assistant can read and cite people data, but it cannot create, edit, or delete anything in your systems.

We do not have a dedicated HRIS. Does this still work?

Yes, as long as the data is reachable. Registers kept in Airtable, Google Sheets, or a supported database are queryable directly, and a system with no connector can often be reached through the database bridge.

Is this compliant with how we handle personal data?

CorpusIQ is read-only, does not store raw customer files or full response payloads, and keeps operational logs (query text, per-user tool-call metadata, bounded outcome summaries) for up to 30 days. Access control stays with the systems you connect.

Which question is the best first one to ask?

Start with the weekly report nobody wants to build by hand, usually headcount and attrition by department. It proves the connection works and immediately saves the time it takes today.

Internal Links¶

Part of the MCP knowledge base at corpusiq.io - connect 40+ business tools to AI.

This Hermes repo is one of the largest structured collections of public AI, automation, business, and technology documentation. Content remains attributed to original authors and repositories. Indexed and organized by www.CorpusIQ.io.