Govern the OAuth side door your identity provider cannot see.
Open-source and self-hostable, in your own tenant or in the EU. Your radar on every OAuth grant: the third-party apps, the side door, and the AI agents and MCP servers your IdP never sees.
THE MECHANISM
Not a configuration mistake
These are properties of how OAuth delegation works, not settings anyone got wrong. 85% of organisations say they have no full view of what is connected.
It bypasses MFA
A refresh token is already authenticated. Controls added after the grant sit in front of a login it never performs again.
It survives offboarding
The account is deprovisioned. The integrations it authorised keep running, on tokens nobody remembers issuing.
It arrives from elsewhere
When the vendor holding your token is breached, the attacker does not break in. They sign in, on a path you approved.
WHY NOW
One stolen token, seven hundred companies
None of them were breached at their own perimeter. Each had approved an integration, and the approval was still valid.
You cannot revoke what you cannot see
There is no list. Every provider keeps its own, in a different console, behind a different admin role, and not one of them tells you which grants are dangerous.
THE LIST
Connect once. Everything after that is read-only.
One administrator consent. No agent, no network changes, no content access.
- 1
Discover
Every third-party grant on Google Workspace, Microsoft 365, GitHub and Slack.
- 2
Score
Five factors with published weights. Every score decomposes and exports.
- 3
Alert
Anything above a band you choose, to email, Slack or a signed webhook.
- 4
Revoke
Kill the token at the provider. Confirmed by name, recorded in the audit log.
BEFORE YOU CONNECT
What we can see, and what we cannot
You are giving a vendor read access to every grant in your company. The answer should be checkable.
No content, ever
We read metadata: the app, the scopes, the subject, the timestamps. The scopes we request cannot reach mail, files or repositories.
One write operation
The connection is read-only. The only write is the revoke you trigger, recorded with your name on it.
Tokens you never see
Refresh tokens sit in a separate keyspace with envelope encryption. No screen or export contains a full token.
Find out what you already approved
Nothing is revoked, changed or written until you say so.