Module · Agents & MCP
The UI is optional.
Most products bolted an assistant onto a screen. LadderHub was built the other way round: the domain layer came first, the Model Context Protocol exposes it, and the screens are one client among several. Ask your assistant or open the app — same permissions, same audit trail.
61 tools, and 23 of them write
Read-only integrations are easy and not very useful. The interesting half is the writes: everything a person does here day to day, an agent can do on their behalf.
- Capture a win, submit it, approve one, link it as evidence for a behavior
- Run a check-in end to end: add topics, comment, submit, review
- Book time off, and rule on a report's — the tools refuse to let a caller count the days themselves
- Set a career target, assess a behavior, move a skill's proficiency, answer a survey, give praise
Every new feature ships its MCP tool in the same pull request. The tool surface does not lag the product, because shipping without one is not how a feature is considered done.
Security is inherited, not reimplemented
An MCP tool here is a thin descriptor over the same domain action the screen calls. There is no second code path with its own idea of who may do what — which is exactly where agent integrations usually leak.
- The same permission check, at the same own / reports / all scope
- The same company isolation — an agent cannot reach another tenant's data
- The same audit log — every write an agent makes is on the record, like any other
- Tools a caller may not use are not listed to them, and lookups that could probe who exists refuse identically whether the person is missing or merely off-limits
Connect it in a minute
LadderHub is its own OAuth 2.1 authorization server, so a compliant client discovers it, registers itself and connects without anybody pasting a secret into a config file.
- Dynamic client registration and PKCE (S256) — no pre-shared credential to leak
- Published authorization-server and protected-resource metadata, plus a JWKS endpoint
- Refresh and revocation, so access ends when you end it
- Or a plain API token, for a script that has no browser to redirect to
Step-by-step setup lives in the MCP integration guide.
The honest part
What agents deliberately cannot do
"The UI is optional" is about the work, not the wiring. Some things are left on the screen on purpose, and it is worth saying which.
Who may do what
Roles, permissions and company setup are not exposed. An integration that can widen its own access is not an integration, it is a privilege escalation waiting to be found.
One-off configuration
The brand accent colour, the platform's AI model provider, prompt templates. Set once by an administrator, with nothing an agent would usefully read or write.
Some HR records, read-only
Leave policies, work schedules, legal entities and pay gap reports are readable over MCP and written on the screen — what somebody is entitled to is an administrative decision, not a chat message.