
What is a MCP for HR AI Agents and AI Tools?
Telling your AI Agents what it can, should, and shouldn’t do
An MCP is a way to guide an AI agent; telling it what it can, should, and shouldn’t do. It acts like the rules, roles, and responsibilities layer for the AI. You’re setting guardrails and instructions.
One of the reasons why an MCP is important is because, at the time I write this (June 2025), most AI Agents in HR sadly fail to deliver what was 'sold' to buyers.
This Fortune article explains how company leaders "describing feeling underwhelmed by agents that failed to deliver" and how AI proofs of concept increasingly fail.
Some of the stuff to be wary of with AI agents today
Before signing an agreement, make sure you test the tools, and you'll find most don't pass. There's some common issues. Test if the AI Agent:
→ Doesn’t know when to escalate
(like answering a complex leave question instead of handing it off to HR)
→ Can’t audit or explain its actions
(HR can't trace what the AI said, to whom, or why)
→ Lacks context re the employee journey
(like treating someone returning from leave the same as a new hire)
→ Makes up policies / legal guidance
(like confidently inventing a “policy” that doesn’t actually exist)
→ Applies the same answer to everyone
(giving identical onboarding tips to managers, interns, and returnees)
→ Doesn’t understand internal tools or channels
(suggests “email HR” when there’s a ticketing system or Slack channel)
These are not edge cases, but pretty common failure points of new AI agents I see being rolled out today.
Hiring a new team member → an AI Agent
When you hire a new HR team member, you'd give them a clear scope and a playbook or detailed instructions. In situations when that new HR team member is unclear on what to do, or how to handle a situation, they should know to ask you first before proceeding!
In a way, and AI Agent needs the same (but probably even more explicit) guidance.
To avoid problems like the ones I mentioned above, it's important to start with real control layers. Part of that it to add a proper Model Context Protocol (or MCP) to guide an AI agent — telling it what it can, should, and shouldn’t do by letting you set guardrails and instructions.
What an MCP usually includes:
- Permissions
→ What data or tools can the AI access? (e.g. BambooHR data, Slack messages, onboarding docs) - Boundaries
→ What shouldn’t it do? (e.g. don’t make payroll changes, don’t send sensitive info) - Intent / Role definition
→ What’s the agent’s purpose? (e.g. “You are an Onboarding Assistant. Your job is to help new hires feel supported, not to answer legal questions.”) - Decision logic or delegation
→ When should the AI act? When should it escalate to a human?
Why it matters for HR:
Without an MCP:
- An AI might overstep (e.g. respond to a termination-related query it shouldn’t).
- It might underperform (e.g. not take obvious next steps after gathering info).
- It might break trust (e.g. answering too confidently on complex topics like leave laws).
With a strong MCP:
- The AI becomes safer, more helpful, and more aligned with HR intent.
- It can operate semi-independently, just like a junior HR team member who has been given clear scope and detailed instructions .
