AI connector — StaffZeno

What it does

The connector lets a recruiter ask their assistant about their own recruiting data — "who are my strongest candidates for the React role?", "what needs my attention today?" — and have it read resumes, score candidates against a job and write those results back into StaffZeno.

It runs on the customer's own Claude or ChatGPT account. StaffZeno holds no AI provider account and no shared API key, so nothing is billed to us and no data passes through an account the customer does not control.

Who can see what

A connection never widens access. It is made per person, and it can reach exactly the records that person can already open when signed in — no more. Permissions are re-read from the database on every single request, so revoking someone's access in StaffZeno takes effect on their next question, not at some later refresh.

Each person connects their own assistant. One person connecting does not connect their colleagues.
A recruiter restricted to their own candidates sees only their own candidates through the connector.
Someone without a module is not even offered the tools for it, and asking for one by name is refused.
Data is scoped to one company account. A token issued for one account can never read another.
An admin can cut off a whole assistant for everyone, in one click, from Settings → Apps.

Authentication

OAuth 2.1 with PKCE. The assistant registers itself through Dynamic Client Registration (RFC 7591); no credentials are issued by hand and none are shared between customers. The person is asked to sign in to StaffZeno and approve the connection once, on a consent screen that names what is being granted.

Two scopes are offered: read-only, and read-and-write. Write is what lets the assistant save a score or a set of screening questions back; it is granted separately and can be declined.

Setting it up

Sign in to StaffZeno and open Settings → Apps.
Copy your account's connector address. It is your own subdomain, like https://your-company.staffzeno.com/mcp.
In Claude or ChatGPT, add a connector and paste that address.
Approve the connection when you are asked to sign in.

Each person on the team repeats this for their own assistant. The address is the same; the access granted is theirs alone.

What the assistant can do

Read

Tool	Returns
`list_job_posts`	Job posts, newest first, with status and whether screening questions exist.
`get_job_post`	One job post including its full description.
`list_candidates`	Candidate search by name, skill or keyword.
`get_candidate`	One candidate's profile.
`get_resume_text`	The text extracted from a candidate's resume.
`list_job_resumes`	Resumes submitted against a job post, with their match scores.
`get_job_resume`	One submitted resume and its scoring detail.
`list_submissions`	Candidate submissions to clients.
`list_interviews`	Scheduled interviews.
`get_interview_transcript`	The transcript of a completed phone screening.
`list_match_runs`	Job match runs and how far each has got.
`get_match_run`	One match run's job description and extracted requirements.
`list_match_run_candidates`	The shortlist queued for scoring in a match run.
`view_match_shortlist`	The ranked results of a match run.
`pending_ai_work`	Counts of what is waiting: unscored resumes, unreviewed calls, unfinished runs.

Write

Only with the write scope granted, and only the fields listed. The assistant cannot delete anything, cannot change money or payroll, and cannot alter another person's records.

Tool	Changes
`save_resume_contact`	Fills in a candidate's name, phone and email read from their resume.
`save_match_score`	Saves a match score and summary against a submitted resume.
`save_match_run_requirements`	Saves the requirements read out of a job description, which queues the shortlist.
`save_match_run_score`	Saves one candidate's score, matched skills and gaps in a match run.
`save_interview_questions`	Saves phone screening questions onto a job post.
`save_interview_evaluation`	Saves the score and summary for a completed screening call.
`add_candidate_note`	Appends a note to a candidate. Never overwrites existing notes.

Turning it off

One person: remove the connector in Claude or ChatGPT, or disconnect from the assistant panel.
Everyone: an admin opens Settings → Apps and presses Disconnect on that app. Every token behind it is revoked immediately.

Data and privacy

Questions and the records needed to answer them go to the customer's own AI provider under their own account, and that provider's terms apply. StaffZeno does not use any of it to train models. The privacy policy sets out what is collected, who it is shared with and how long it is kept.

Support

Write to noreply@staffzeno.com — a person reads it.