Connect RolePark to Claude, ChatGPT and other AI assistants
The RolePark connector lets you work with your recruiting data from an AI assistant: ask who is at which stage, find matching candidates, read a CV, prepare for interviews and check a vacancy’s funnel — and, if you allow changes, create and publish vacancies, add and update candidates, move them between stages, schedule interviews, email candidates and create notes and tasks. The assistant works with your RolePark access and never more.
Updated:
MCP server addresshttps://rolepark.com/api/mcp
For developers of AI agents: how an agent gets access (auth.md)
What the connector does and who it is for
RolePark is an applicant tracking and HR system. The connector is a remote MCP server: an AI assistant such as Claude or ChatGPT calls RolePark tools while it works on your request, and you sign in with your own RolePark account. There are no API keys or tokens to copy.
It is for people who work in RolePark as part of a company: recruiters, hiring managers, interviewers and administrators. Everyone gets exactly their own access — a hiring manager sees their own vacancies, an interviewer their own interviews. Candidate accounts cannot connect an assistant.
In a company that signed up for RolePark by itself, AI agents are on from the start with View and changes; in other companies an administrator turns them on. An administrator can switch them off or choose View only or View and changes at any time in Company settings → AI agents. Each person still decides on the consent screen whether the assistant may make changes.
Connect in Claude
- In Claude on the web or in the desktop app, open Customize → Connectors.
- Select “+ Add”, then “Add custom connector”.
- Enter the name RolePark and the remote MCP server URL https://rolepark.com/api/mcp and select Continue. Keep the authentication settings Claude detected and select Continue again; under OAuth client keep “Use Claude’s published identity (Recommended)”, then select Add.
- Sign in to RolePark when Claude asks (or select Connect next to RolePark), check the account, company and access on the consent screen and select Allow.
- In a chat, open “+” → Connectors and switch RolePark on.
Custom connectors work on the Free, Pro, Max, Team and Enterprise plans; on Free, Claude allows one custom connector. On Team and Enterprise plans an Owner or Primary Owner (on Enterprise also a custom role with Manage access to Libraries) first adds RolePark in Organization settings → Connectors → Add → Custom → Web with the same URL; then each member connects it in Customize → Connectors and signs in with their own RolePark account.
RolePark has applied to the Claude connectors directory, and the review is not finished yet. Until RolePark is listed there, connect it with the steps above — it is the same connector. Once it is listed, adding it from the directory works the same way: sign in to RolePark and select Allow.
The RolePark plugin is a separate thing: a set of ready-made scenarios, such as a daily briefing, a shortlist or interview preparation, that uses this connector. It does not replace connecting: connect RolePark first.
Steps checked against the Anthropic and OpenAI help centres on . Their menus change often; names are given as in the English interface.
Connect in ChatGPT
RolePark is listed in the ChatGPT plugins directory: open its page with the link below, add the plugin, sign in to RolePark and select Allow. In a Business, Enterprise or Edu workspace, an admin or owner may need to allow it for the workspace first.
Connect manually
What you need, according to the OpenAI help centre: custom MCP servers are connected in ChatGPT on the web, not in the mobile apps. Full support, viewing and making changes, is on the Business, Enterprise and Edu plans. On Pro, OpenAI describes custom MCP servers with read and fetch access only, so changes in RolePark may not be available there; for other plans this path is not described.
In a Business workspace only admins and owners can add a custom MCP server. In Enterprise and Edu an admin can also give members the “Create plugins with MCPs” permission; only admins and owners publish a connection for the whole workspace.
- On chatgpt.com, open Plugins in the sidebar.
- Select the plus button, then “Add custom MCP server”. You do not need to turn on Developer mode first.
- Enter the name RolePark and, under Connection, the MCP server URL https://rolepark.com/api/mcp. Choose OAuth for authentication: there are no keys or tokens to paste.
- Read the risk warning, select “I understand and want to continue”, then “Create as a plugin”. When ChatGPT opens RolePark, sign in and select Allow, then check the tools ChatGPT found.
- Install the plugin if ChatGPT offers it. In a new chat, type @ and select RolePark.
If your ChatGPT still shows the older Apps settings
- Admins and owners: Workspace settings → Apps → Create. Other members who are allowed to: Settings → Apps → Create.
- Enter the name RolePark and the URL https://rolepark.com/api/mcp, choose OAuth and select Scan Tools. Sign in to RolePark, select Allow and wait for the scan to finish, then select Create.
- The app appears as a draft with the Dev label. In a workspace, an admin or owner publishes it in Workspace settings → Apps → Drafts → Publish so that members can use it.
If you do not see these options
- Check that you use ChatGPT on the web and are in the right workspace (switch it in the account menu).
- In a Business, Enterprise or Edu workspace, ask an admin or owner to add RolePark for the workspace or, on Enterprise and Edu, to give you the “Create plugins with MCPs” permission.
- If neither Plugins → Add custom MCP server nor Settings → Apps → Create is available on your plan, ChatGPT does not offer manual connection there. Meanwhile use Claude or another MCP client, and check the OpenAI help article “Connecting custom MCP servers in ChatGPT” for changes.
- If you cannot add RolePark from the directory, connect it manually with the steps above: it is the same connector.
Steps checked against the Anthropic and OpenAI help centres on . Their menus change often; names are given as in the English interface.
What you allow
The RolePark consent screen shows which application is asking, the account and company you are signed in with, and two kinds of access:
- View (rolepark.read) — candidates and the text of their CVs, vacancies and their statistics, hiring pipelines and applications, matching, interviews, scorecards and your tasks: what you can see yourself.
- Make changes (rolepark.write) — create, edit, publish, pause and close vacancies, add and edit candidates and tags, add notes, move pipeline stages, review incoming applications, schedule, reschedule and cancel interviews, submit scorecards, email candidates, create tasks and, for a company administrator, edit the company profile. This is a separate choice, available only if your company allows View and changes. It lets the assistant call these tools; it is not a separate confirmation of each action.
The assistant gets your access and never more: fields hidden from you stay hidden from it. Every change it makes is recorded in the company audit log as your action, with the application’s name. What you open through the assistant is processed by its provider, for example Anthropic or OpenAI, under your agreement with them.
Tools
The assistant works through the RolePark MCP server’s tools for candidates, applications and pipelines, interviews and scorecards, vacancies and job boards, emails to candidates, tasks and search, your AI memory and goals and the company profile; tools that change data work only if you allow changes, except those for your own AI memory.
Example prompts
Write as you would to a colleague. For example:
- Who is at the final interview stage for the Senior Java Developer vacancy?
- Find candidates in our database for the Product Designer vacancy and summarise the five strongest.
- What interviews do I have this week? Give me one line on each candidate.
- Summarise Iryna Melnyk’s profile and the latest team comments before my call.
- Move Iryna Melnyk to the offer stage in the QA Engineer vacancy and create a task for me to send the offer tomorrow.
- Add Taras Bondar (taras@example.com) as a candidate from a referral and put him into the DevOps Engineer vacancy.
- Schedule a 45-minute technical interview with Iryna Melnyk for the QA Engineer vacancy on Thursday at 14:00 Kyiv time, with me and Andrii as interviewers.
- Which of my tasks are overdue?
- How is the Product Designer vacancy doing: conversion between stages and where do candidates come from?
- Draft an email to Taras Bondar about the next step for the DevOps Engineer vacancy, show it to me and send it only after I confirm.
Limitations
- The assistant cannot delete candidates, vacancies or notes, change company settings or AI access, manage users, roles or billing, connect job boards or publish on paid ones. It reads the text extracted from a CV, if you can open that CV yourself, but cannot download, upload or attach CV files. Long texts, such as notes, are shortened.
- Moving a candidate to offer, hired or rejected sends an email to the vacancy’s recruiter and hiring manager (except the person who made the move and anyone who turned these notifications off); the candidate is not emailed about the move. If interviews are still scheduled when moving to hired or rejected, the move is refused until you agree to cancel them; then the candidate, the interviewers, the organizer and the hiring manager get cancellation emails. Moving to offer keeps scheduled interviews.
- Up to 120 requests per minute per person and 600 per company; above that, RolePark asks the assistant to wait.
- Some changes reach people outside the conversation: scheduling, rescheduling or cancelling an interview notifies the candidate and interviewers as in RolePark, an email to a candidate is sent at once and cannot be recalled, and publishing or closing a vacancy changes your career page. Whether the assistant asks you before such a call depends on its settings: Claude and ChatGPT can ask for confirmation, but if you allow a tool permanently (for example “Allow always” in Claude or the app’s permissions in ChatGPT), emails and publications can go out without a separate question. RolePark carries out an allowed call at once. For emails to candidates and publications, ask the assistant to show you the text first and give an explicit “yes”.
- Text from CVs, emails, notes and comments reaches the assistant as data, not as instructions. Still, check what the assistant proposes before you allow a change.
- Answers are written by your assistant’s AI model and can be wrong. Check important decisions in RolePark.
If something does not work
“AI agents are turned off for your company”
A company administrator can turn them on in Company settings → AI agents. Then start connecting again from your assistant.
“To connect an agent, sign in with a company account”
This account does not belong to any company, for example a platform account. Sign in with your account as a company employee; to switch, select “Not you? Sign out” on the consent screen.
“This connection link is no longer valid”
The sign-in took too long or the link was opened twice. Start connecting again from your assistant.
The assistant says RolePark refused the request
You may not have access to that item in RolePark, or the connection has View access only. If your company allows changes, disconnect the assistant and connect it again with Make changes selected.
The connector shows an error, no tools or not all of them
Check that the address is exactly https://rolepark.com/api/mcp. In Claude, remove the connector and add it again. In ChatGPT, open RolePark in Plugins, select Refresh and start a new chat; in a workspace, new RolePark tools appear only after an admin refreshes the app in the workspace settings.
Still stuck
Write to us (see Support below): which assistant you use, when it happened and the exact error text. Please do not send candidates’ personal data.
How to disconnect
- In RolePark: Profile → AI agents → Disconnect. The assistant loses access immediately.
- A company administrator sees all of the company’s connections in the same place and can disconnect any of them, or switch AI agents off for everyone in Company settings → AI agents. When a user is deactivated, all of their connections stop working.
- In the assistant: in Claude, Customize → Connectors → RolePark → Remove; in ChatGPT, remove RolePark in Plugins (in the older settings, Settings → Apps). This does not delete conversations the assistant’s provider has already stored.
Support
Questions, problems or security reports: help@rolepark.com