notanATS
notanATS is screening infrastructure. It holds the hiring manager's definition of the work, screens applicants against it and moves the right people to the interview. Your own agents and apps can tap into that process through the API. The screening layer becomes part of the system you already operate, rather than another queue someone has to watch.
Hosted MCP Server
npx add-mcp 'https://notanats.com/mcp'Installs into Claude Code, Codex, Cursor and more
Documentation
COMPARE!your process vs notanATS →
01
Define
The job description becomes a draft screening model of the role.
02
Calibrate
You correct it in a short chat and lock it. Your definition, in your words.
03
Screen
Every applicant is screened in a chat, against the locked model.
04
Review
You see each applicant's standing, dimension by dimension.
05
Interview
You interview the right people. That part stays human.
Screening is where hiring slows down.
The job gets posted, and applications arrive faster than anyone can read them. Screening them one by one takes the manager's week. Or it gets rushed — and then the wrong people reach the interview while the right ones wait in the pile.
The problem is not effort. It is that the manager's definition of the right applicant lives in the manager's head, and it has to be applied to every application, by hand, every time.
notanATS takes that definition out of the manager's head and puts it to work.
You define the work. The framework holds the definition.
When a position is created, notanATS reads the job description and drafts a model of the role — up to five dimensions. Each dimension carries three things:
| The work Shown to applicants | What the job actually involves, in plain terms. Applicants see this before they are screened, so they know what they are being screened on. |
|---|---|
| What good looks like Private — never shown | Your scoring rubric: what a strong answer proves, and what a weak one gives away. It is never shown to applicants — and the screening chat is built so it cannot see it either. More on that under trust & fairness. |
| What capability looks like Shown to applicants | The concrete signals you would accept as proof — the answers, examples, and behaviours that show someone can actually do this work. |
A short calibration chat then walks you through the draft. You confirm, correct, or cut each dimension, mark the ones that are must-haves, and lock the model. From that moment, every applicant for the role is screened against the same definition: yours.
01 Job description You add the role, its work and duties. 02 Calibration You correct the draft and lock the model. Rubric stays private 03 Screening chat Applicants answer questions from it. Shown to applicants 04 The pool Scores on every dimension, one view. Shared at close 05 First interview You meet the right people, in person.
Shown to applicants the applicant sees it — their standing, when the position closes Private — never shown the rubric never reaches the applicant
The rubric is set during calibration and never crosses into the chat. The screening side is built so it cannot see it — the questions travel, the answers behind them do not.
A chat, not a form. Your rubric, not a keyword scan.
Applicants apply, add a resume, pick a time, and get a private link — there is no account to create. At their time, the screening opens: a typed conversation about the work itself, built from your locked model.
Their name and email are masked out of the record the AI screens and scores. The questions follow your dimensions, and the conversation digs into the answers the way a good screener would — including the things that look right on paper but fail in practice.
When the chat ends, the applicant's evidence is matched against your rubric. Each dimension gets a rating, and the ratings combine into a score out of 100.
Typed, up to an hour
A written conversation — no camera, no voice recording. The applicant can book a time or start right away.
Built from your locked model
Every question traces back to a dimension you calibrated — including the anti-patterns, the things that look right on paper but fail in practice.
You review people, not piles.
Every scored applicant lands in one pool view, with their standing on each dimension — not just a total. You advance people one by one, or turn on automatic advancement and let the top of the pool move in batches. Either way, the interview is booked inside notanATS.
Depot Dispatcher — poolScored against the locked model
| Applicant | Route planning | Pressure | Record keeping | Score |
|---|---|---|---|---|
| Applicant 014 | Strong fit | Strong fit | Fit | 86 |
| Applicant 007 | Strong fit | Fit | PartialMust-have missed | 74 |
| Applicant 021 | Fit | Partial | Fit | 68 |
An example pool, written for this page. Note the flag: an applicant who misses a must-have is marked, not hidden. They stay in the pool, and you can still advance them by hand.
Dimension by dimension
A total score hides where someone is weak. The pool shows every dimension separately, so the trade-offs are yours to make.
Nothing buried
Must-have misses and suspected automation are flagged on the record, in the open — never quietly filtered out.
The call is yours
notanATS screens and scores. Advancing an applicant to interview is a person's decision — unless you turn on automatic advancement for the top of the pool.
Our competitor could be your current process. Compare it and see.
Run one live role through your current process and through notanATS at the same time — one link deals each applicant to one side or the other — and see which side reaches the first interview, and the ones after it, faster. How the test runs →
A framework your agent can operate.
notanATS is built to be run by the agent you already work with. Connected with your organisation's API key, your agent moves the work along — inside limits that keep the judgment with you.
Your agent can
- Create a position from the job description
- Hand you the calibration link so you can define the work
- Read the pool — scores and per-dimension standing
- Advance the applicants you choose to interview
Your agent cannot
- Calibrate. Defining the work is the manager's job, done by the manager, signed in. No exceptions.
- Screen anyone. The agent is a control channel. notanATS does the screening.
Works with the agent you already use.
ChatGPT 
Claude 
Google Gemini 
Perplexity
Any agent that works over an API or MCP
The agent moves the work along. The definition is yours. The screening is the framework's. The interview decision is yours.
Nobody is left without an answer.
Screening usually ends in silence. Here, when a position closes, applicants can see where they stood in the pool — on each dimension they were screened on — where their organisation's transparency settings allow it. Not a single overall rank, and not a personal debrief. The same definition they were screened against, shown back to them.
It is a fairer ending, and it costs the manager nothing. The feedback is built from the screening itself.
Screened on the work
A conversation about the actual job — in the applicant's own words, not a keyword match against their resume.
No account, no password
A private link, checked against their inbox, that opens only inside their booked screening window.
Not tracked
Analytics never run on applicant pages. The screening is not a marketing surface.
Define the work once. Meet the right applicants.
Fairness is monitored while you go: selection rates are checked with the EEOC four-fifths rule, and drift gets flagged. How that works.