ChatGPT ATS integration: what recruiters need to know about MCP

A ChatGPT ATS integration is a connection that lets ChatGPT read from and act on your hiring system, so a recruiter can ask for a pipeline or a candidate's results in plain language. Today that connection runs over MCP, the Model Context Protocol. This post is for recruiters and hiring teams, not for candidates trying to get a resume past an applicant tracking system.

Our view: connecting the assistant is the easy part. The useful question is what the assistant is allowed to do once it's in, and who is still making the decisions.

Key Takeaways

Our position: a ChatGPT ATS integration is an MCP question, and what matters most is what the assistant can and can't do once it's connected.
MCP is an open standard that lets assistants like ChatGPT and Claude connect to outside systems. It replaced the old copy-and-paste workflow.
There are three routes: your ATS ships an MCP server, your evaluation tool ships one, or you wait and work from exports.
Before switching it on, check sign-in, organization scope, excluded actions, the ChatGPT Developer mode warning, and who can add connectors.
An assistant can fetch a report card. A person still reads the evidence and makes the call.

ChatGPT ATS integration: what it means for a recruiter

Searches for this term mostly return resume tips, because candidates want to know if ChatGPT can help them beat applicant tracking. That's a different problem. Here we mean the employer side: using ChatGPT for recruiters' daily work, connected directly to the system where jobs, candidates and results live.

The old way was copy and paste. You exported a list, pasted it into a chat, asked for a summary, and pasted the answer back somewhere else. It worked, but the assistant never saw live data and couldn't act on anything.

A real integration changes that. The assistant can look up the current pipeline, pull a candidate's results, or create a job, all inside a conversation and limited to what your account is allowed to see.

How an assistant connects to your hiring system: MCP in plain English

The Model Context Protocol describes itself as an open-source standard for connecting AI applications to external systems. It names Claude and ChatGPT as examples, and compares MCP to a USB-C port for AI applications. One plug shape, many devices.

In practice, the hiring tool runs an MCP server. Your assistant connects to it, discovers what it can do, and calls those tools when you ask. For remote servers, the MCP authorization specification covers sign-in, which is how the server knows who's asking.

That's why MCP for recruiters matters. It works the same whether you connect ChatGPT or set up an ATS Claude integration, so you aren't locked into one assistant.

Three ways to get ChatGPT or Claude into your recruiting stack

If you want to connect ChatGPT to ATS data or evaluation results, you have three realistic routes today.

Your ATS ships an MCP server. Some do. Stardex, for example, has published a walkthrough on connecting ChatGPT directly to its ATS. If your ATS offers one, this is the most direct route for pipeline data.
Your evaluation layer ships one. This is where Expert Hire sits. The Expert Hire MCP server connects Claude or ChatGPT over OAuth, and nearly everything you can do in the platform is available through it, so you can work on jobs, candidates, interviews and report cards in your own words. It connects the assistant to Expert Hire, not to your ATS.
Nothing ships one yet. Then you wait, or keep using exports, and ask your vendors about their plans.

So which HR tools have MCP integration? It's a small but growing group, and the list keeps changing. Ask each vendor directly what their server exposes, what it excludes, and whether it's generally available or in early access.

What to check before you turn it on

MCP servers for recruiting touch candidate data, so treat setup like any other access decision. Here's the checklist we'd use.

Sign-in that runs as the user. OAuth means the assistant acts as the person who connected it, not as a shared admin account.
Scope to your organization. The assistant should see what your own account already sees, and nothing else.
Excluded actions. Ask what the server deliberately leaves out. Destructive, financial and admin actions should usually be off the table.
The ChatGPT side. Connecting a custom MCP server needs Developer mode and a ChatGPT plan that supports custom apps. ChatGPT warns that Developer mode allows unverified connectors that could modify or erase data permanently, and that memory is disabled.
Who can add connectors. On Claude Team plans, the Anthropic help guide to custom connectors says only Owners and Primary Owners can add them. It also warns that a malicious server can hide instructions meant to make Claude do things you didn't intend, so only connect servers you trust.

Expert Hire's server is built around these points. It connects over OAuth, every call runs as you and stays scoped to your organization, and financial actions, team-admin changes and destructive operations are excluded by design. It's in early access, and it unlocks on the Growth plan ($499 a month per workspace), with higher limits on Scale.

The guide to connecting Expert Hire to ChatGPT walks through the exact screens. Claude works as a custom connector with the same server URL and the same OAuth sign-in.

What you can ask once it's connected

Start with a read-only request. It's the safest way to confirm the connection works and that the scope is what you expect.

With the Expert Hire server, the MCP page lists four areas: jobs and rounds, candidates and resumes, interviews and scheduling, and reports and analytics. Here are requests that fit those areas:

"Show me the pipeline for the senior backend role."
"Pull the report card for the last three candidates on the data analyst job."
"Which interviews are scheduled for this week?"
"Create a job for a product designer and add an AI interview round."
"Shortlist the two strongest candidates on the support role and tell me why."

That last one is where judgment comes in. The assistant can gather the evidence and propose a shortlist, but someone on your team should open the report cards before anything moves.

If you also run Greenhouse, that's a separate connection. The integrations page lists Greenhouse as available today in early access, and our post on AI interview ATS integration explains how results flow back to an ATS.

Where an assistant should stop

An assistant that reads a report card isn't making a hiring decision. It's fetching and summarizing evidence, and summaries can miss things. Read the transcript and the scores behind any candidate the assistant flags.

Automation still has a place. Expert Hire workflows can auto-shortlist or auto-reject against a threshold you choose, but both are off until you turn them on. When they're on, every decision has a transcript and a scored report behind it, so there's no black-box pass/fail.

Our post on what a hiring copilot should and shouldn't do covers where autonomy stops. The short version: let the assistant do the fetching, and keep people on the decisions.

Frequently asked questions

Can you integrate ChatGPT with an ATS?

Yes, if the system you want to reach runs an MCP server. You turn on Developer mode in ChatGPT, add the server as a custom app, and sign in with OAuth. Some ATS vendors ship their own servers, and evaluation tools like Expert Hire ship one for their own data.

Does Claude work with an ATS the same way?

Yes. MCP is an open standard, so the same server can work with ChatGPT and Claude. In Claude, you add it as a custom connector. On Team plans, only Owners and Primary Owners can add connectors, so involve your admin early.

Which HR and recruiting tools have an MCP integration?

A growing but still small group, and many are in early access. Ask each vendor what their server exposes, what it excludes, and how sign-in works. Expert Hire's MCP server works with ChatGPT and Claude and unlocks on the Growth plan, with higher limits on Scale.

Is it safe to connect ChatGPT to candidate data?

It can be, with the right setup. Use OAuth so the assistant acts as you, confirm the scope is limited to your organization, check which actions are excluded, and only connect servers you trust. Start with read-only requests while your team learns how it behaves.

Can employers tell if a resume was written by AI?

Not reliably, and it's rarely the right question. A polished resume can still describe real work. What matters is whether the candidate can explain their claims in depth, which our guide on how to verify skills on a resume covers in detail.

The bottom line

A ChatGPT ATS integration is now a practical thing, not a future plugin. Connect it over MCP, check the scope and the excluded actions, start read-only, and keep a person on every decision the assistant helps with.

If you want ChatGPT or Claude working against your interviews and report cards, see how the Expert Hire MCP server connects.

By TK, Growth at Expert Hire. Last updated October 13, 2026. Reviewed by Anand Suresh, CPO at Expert Hire.