An HR MCP server: your hiring pipeline and HR tasks, from Claude or ChatGPT

HRMS exposes 14 actions to a connected assistant. They cover two parts of the tool — the workflow tasks waiting on you, and hiring — and they are the whole list:

• List the workflow tasks assigned to you, such as a probation review, and open one to read what it asks.
• Complete your step on a task, with the answers you give the assistant.
• List the open jobs.
• List the candidates on a job and see which stage each has reached.
• Add a candidate to a job.
• Move a candidate to another stage of the pipeline.

That is narrower than HRMS itself, on purpose. Hiring and task follow-up are the jobs people do from wherever they happen to be; the records about a person stay in the tool. The examples people actually type, tool by tool, are on the Orbiz MCP page.

A referral arrives in your inbox on a Tuesday. Instead of opening HRMS, finding the job and filling the form, you tell the assistant which job the person is for and where they came from, and it adds the candidate. Before the weekly hiring meeting you ask who is in the interview stage for each open role and get one answer instead of four screens. After the meeting you ask it to move two people forward.

The same goes for tasks. HRMS workflows send each step of a process to the person who owns it — a manager’s probation evaluation, for instance. An assistant can tell you what is waiting, read you the form and submit your answers. The judgement stays yours: AI screening is a separate HRMS feature that scores a résumé against the job inside the tool, and a connected assistant does not replace it or produce those scores.

Connecting an assistant opens an Orbiz consent page where you are signed in as yourself. It names the assistant, your organisation and your account, and lists access tool by tool, with reading and changing as separate ticks. For HRMS that means you can let an assistant list jobs and candidates without letting it add or move anyone. You can untick anything before you approve, and cancelling grants nothing.

The page only offers what your role could use. A member whose role has no hiring permissions is not offered hiring access to hand over, and the consent page says so in one line: the assistant can only see and do what you can on the web, in the areas you allow.

Every member has a list of their own connections in Settings → Connected AI apps: the assistant, what it was allowed, when it connected and when it was last used. Below it is an activity log of each action with its outcome — done, not allowed or failed — and the time. Admins get the same two lists for the whole organisation, with the member’s name on each row and a filter by member.

Two controls sit beside those lists. An admin can disconnect any single connection, and can switch AI assistants off for the organisation, which stops every assistant at once and blocks new connections until it is switched back on. The wider platform controls are on the security page.

There is nothing to install and no key to generate. In Orbiz, open Settings → Connected AI apps: it shows your organisation’s connection address with a copy button and the steps for each assistant. Add the address as a custom connector in Claude or ChatGPT, sign in to Orbiz when the assistant sends you there, and tick HRMS. The address is shown only inside the product, which is why it is not printed here.