ChatGPT MCP for leave, time and HR
Connect ChatGPT to SquadBear over MCP. Handle leave, time and team health in plain language, on a curated catalog, with your own permissions.
SquadBear's MCP server is the same product ChatGPT, Claude and other MCP clients talk to. ChatGPT uses its own URL: a curated subset of the catalog, not a second permission model.
Curated catalog
ChatGPT talks to /mcp/chatgpt. Everyday leave, time and health. Not policy edits or erasure.
Acts as you
The connection is your account. An employee's ChatGPT cannot approve anyone else's leave.
Same audit trail
Every tool call is recorded next to UI clicks, at Workspace settings → Audit log.
No extra SquadBear fee
MCP is on every SquadBear plan, including free. Your ChatGPT plan still has to support custom connectors.
What you can do in ChatGPT
Check and book leave, log time, pull reports, run a health check, manage records, chase a handbook signature. Answers come from your real data rather than a guess.
Leave-policy edits, workspace settings, erasure and process starts are on the default /mcp endpoint used by Claude, Grok and Gemini - not on this one.
What it does not do: emotion or mood inference, candidate screening, ranking people for hire / fire / promote, or automatic people decisions off a score.
Ask it to run get_context first. It should reply with your identity, workspace and connection permissions.
How to connect ChatGPT
- Your ChatGPT endpoint is
https://app.squadbear.com/mcp/chatgpt. The in-app AI agents page also shows it. - In ChatGPT, open Settings → Connectors and add a custom MCP connector (Plus, Pro, Team or Enterprise). Paste that URL. Not the Claude one.
- Sign in to SquadBear when prompted, then approve the scopes the connection may use. The default covers leave and the directory. Time and health need an extra tick, and those modules have to be on.
- Enable the SquadBear connector in your conversation and start asking.
Because the permission model is shared, Connect Claude to SquadBear is still the canonical reference for scopes. The endpoint in that guide is Claude's. ChatGPT's is /mcp/chatgpt.
Permissions and audit
A connected assistant acts as you, with your role's permissions and no more. The scopes you approve are frozen into the connection; widen them by reauthorizing. Every action is recorded in the same audit trail as the web app and REST API.
Which assistant, which URL
| Assistant | Endpoint | How you add it |
|---|---|---|
| Claude | /mcp | claude.ai or Desktop. Add to Claude prefills the connector. A free account can add one. |
| Claude Code | /mcp | One terminal command, then /mcp Authenticate. |
| ChatGPT | /mcp/chatgpt | Curated catalog. Plus, Pro, Team or Enterprise. |
| Grok | /mcp | grok.com connectors, or grok mcp add in Grok Build. |
| Gemini CLI | /mcp | gemini mcp add, then /mcp auth. |
| Cursor | /mcp | Add to Cursor on the MCP page, or an HTTP server in Cursor settings. |
| Codex | /mcp | ~/.codex/config.toml, then codex mcp login. |
A token minted for /mcp is rejected on /mcp/chatgpt, and the reverse. Same permissions and audit trail either way.
Prompts to try
Related
- Claude · Grok · Gemini CLI · Claude Code
- Connect an agent to SquadBear
- The developer API
- Security, data residency and GDPR
Frequently asked questions
Does SquadBear work with ChatGPT?
Yes. ChatGPT on a plan that supports custom MCP connectors can operate SquadBear. It talks to https://app.squadbear.com/mcp/chatgpt, a curated subset of the same catalog Claude uses.
Is it the same URL as Claude?
No. Claude uses https://app.squadbear.com/mcp. ChatGPT uses https://app.squadbear.com/mcp/chatgpt. A token is bound to one endpoint. A token minted for /mcp is rejected on /mcp/chatgpt, and the reverse.
Can ChatGPT see data I can't?
No. The connection resolves to your account and carries your role's permissions. An assistant can only ever do what you could do yourself in the web app.
Does ChatGPT decide employment outcomes?
No. It operates SquadBear tools under your permissions: draft, fetch and file. No candidate screening, no performance scoring for promotion or termination, no emotion inference. Humans stay in command of people decisions.
Can ChatGPT change a leave policy or erase a person?
Not on the ChatGPT endpoint. Leave-policy edits, workspace settings, erasure and process starts live on the default /mcp catalog, not the curated ChatGPT subset.
Do I need to write any code?
No. Connecting an assistant over MCP is a chat-based setup. For scripts and backend jobs, use the REST API with an sbk_ API key instead.
Start free in minutes
Every module on every plan, including free for up to 5 people. Connect your assistant and try it on your own data.
Start free
