Model Context Protocol
Your AI can already read your ATS. Now it can tell you who’s good.
The ProboTalent MCP Server connects Claude, ChatGPT, or Grok to your hiring data. Ask in plain language, get answers grounded in assessment results and verified candidate evidence — under your own permissions, read-only, and logged.
Already a ProboTalent customer? Ask your Executive Admin to enable AI / MCP Access, then connect from inside ProboTalent.
Which of my finalists for Warehouse Lead has the strongest evidence — and what’s still missing on each?
Three finalists are on Warehouse Lead. Here’s the evidence on file for each, and what’s still open:
| Evidence | D. Okafor | M. Reyes | J. Lind |
|---|---|---|---|
| Phase 2 result | Above threshold | Above threshold | At threshold |
| Culture fit | High | High | Moderate |
| Manager references | 2 of 2 complete | 1 of 2 complete | 2 of 2 complete |
| ID verification | Passed | Passed | Not started |
| Still missing | Nothing | One manager reference | ID verification |
D. Okafor has the most complete evidence. Want the full evidence pack on any of them?
Source: ProboTalent · read as Adrienne D. · 09:42 SAST · read-only
What it is
Your AI assistant, connected to ProboTalent.
MCP — the Model Context Protocol — is the open standard Claude, ChatGPT, and Grok use to connect to the software you already use. The ProboTalent MCP Server is the first-party way to connect them to ProboTalent. Other hiring platforms have opened their doors to AI too; what an assistant finds inside is plumbing — records, stages, notes. Here it reads assessment results, verified work history, and reference feedback, so it can answer “who is ready, and what’s missing” instead of listing names by the date they applied.
Ask in plain language.
No commands, no syntax, no exports.
Answers from real evidence.
Assessment results, verified work history, reference feedback — not a summary of a résumé pile.
Read-only, under your permissions.
It can look, count, compare, and report. It cannot change a thing.
What you can ask
Ask it like you’d ask a colleague.
- “Which of my finalists has the strongest evidence — and what's still missing on each?”
- “Who's been sitting in Phase 2 for over a week with a completed report nobody has opened?”
- “Before I post this job, who already in our talent base fits it?”
- “Why was Marcus marked Not a Match, who set it, and when?”
- “Has every finalist passed ID verification and got both manager references in?”
- “What did Maria's references say about how she handles pressure?”
- “Where did last quarter's hires actually come from — our own talent base or new sourcing?”
- “What's waiting on me today?”
Every answer says when it was read and what filters applied, and links straight back into ProboTalent so you can act on it there.
How it works
Connected in minutes. Governed from day one.
Everything happens inside ProboTalent. Nothing to download, nothing to set up on this website.
Your admin switches it on.
AI access is off by default for every organization. An Executive Admin enables it under Organization Settings → Integrations → AI / MCP Access and chooses which categories of data are in scope: jobs, candidates, scores and reports, references and evidence, and analytics. Anything outside an enabled category is invisible to every connected assistant, and switching a category off takes effect within a minute.
You connect your assistant from inside ProboTalent.
Once access is enabled, the connection details and step-by-step instructions for Claude, ChatGPT, and Grok are in ProboTalent, under your profile. Follow them, and sign in with your ProboTalent account when your assistant asks. That sign-in is the whole permission model: your assistant becomes you, with exactly your roles.
You ask.
Type your question in your assistant. It reads live data from ProboTalent, writes up the answer, and links you back into the platform.
Customers: full setup guide in the Help Center
Security and privacy
Built the way your security team would build it.
We designed the MCP Server by asking what an AI should never be able to do with hiring data, and enforcing the answers once, in the server, for every tool.
It sees only what you see
Every request runs through ProboTalent's roles engine as the signed-in user. A record you can't open in the app is reported to your AI as not existing — not as forbidden — so nothing leaks.
Read-only, by design
No tool creates, edits, moves, archives, sends, publishes, or deletes anything. Your AI can look, count, compare, and report.
Certain data never leaves
EEO/EEOC self-identification data, the wording of proprietary assessment questions, consent records, Social Security numbers and dates of birth, identity documents, credentials, and other users' task queues are blocked at the server — in results, errors, and logs alike.
Contact details stay masked
Candidate email addresses and phone numbers are returned masked in every response.
Every request is logged
Who asked, from which AI app, which data was read, and what came back — with contact details and free text scrubbed. Every user can see and disconnect their own Connected AI apps; deactivating a user ends their connections immediately.
No bulk export, no cached data
Results are paged and capped, files come back as time-limited links rather than streamed into the AI, every read hits live data, and rate limits stop a runaway session.
ProboTalent’s results come from assessments designed by Industrial-Organizational psychologists — not from an AI’s opinion of a candidate. Your assistant quotes the stored result, explains what your thresholds mean, and says what evidence is still missing. It never produces a score of its own, never recommends advancing or rejecting anyone, and every decision stays with a person.
- SOC 2 Type II
- ISO 27001
- GDPR compliant
- Open MCP standard
FAQ
Questions we get asked first.
- Which AI assistants can I use?
- Claude, ChatGPT, and Grok are tested and supported. Because MCP is an open standard, other assistants that support remote MCP servers with sign-in can also connect, but we only stand behind the three we test.
- Can my AI change anything in ProboTalent?
- No. This release is read-only. If you ask your assistant to do something, it will tell you where in ProboTalent to do it.
- Can my AI see anything I can't?
- No. Every request runs under your own ProboTalent roles and permissions. If you can't open a record in the web app, your assistant is simply told it doesn't exist.
- Where do I connect my assistant?
- Inside ProboTalent, not from this website. Once your Executive Admin has enabled AI / MCP Access, the connection instructions appear under your profile in the application.
- Does ProboTalent store my conversation with the AI?
- No. ProboTalent logs each request — who, which AI app, which data was read — not the conversation itself. Your conversation lives with your AI provider under their terms.
- Does the AI make or influence hiring decisions?
- No. It reads and explains results ProboTalent already holds. It never produces a score or recommends advancing or rejecting a candidate.
- Will it be able to take actions later?
- Possibly, and only with confirmation, attribution, and audit designed in from the start. We shipped read-only first so teams can adopt it without a governance debate.
Give your AI something worth reading.
If you use ProboTalent, ask your Executive Admin to enable AI / MCP Access today. If you don’t yet, see it live on one of your real roles.