Agent lifecycle management

Question: How do we design the onboarding and offboarding process for agents, so that new agents gain access quickly and decommissioned agents are reliably blocked?

Approach

Keycloak handles lifecycle management as the identity provider (step 3).

• Onboarding: a new agent is created as a client/identity in Keycloak and given the appropriate scopes and roles. Access follows from the issued token – the MCP server itself needs no change.
• Offboarding: when an agent is decommissioned, it is deactivated in Keycloak or its tokens are invalidated. The MCP server validates tokens against Keycloak and thus reliably denies access.

Benefit

Permissions and lifecycle are managed centrally in Keycloak, not scattered across the MCP server. This keeps onboarding and offboarding fast and traceable, and fits with Security & access control.

Emergency / incident response

The lifecycle is also where an emergency concept for revoking access applies:

• Immediate blocking (Kill Switch): if an agent is compromised or runs amok, it is deactivated in Keycloak or its tokens are revoked. How fast this takes effect depends on the token lifetime: with long-lived tokens the block takes effect only after a delay. Short token TTLs (or token introspection per request) shorten the window – a deliberate trade-off between response time and load.
• Traceability during an incident: for the follow-up, the Audit Layer applies – lifecycle (who was active when) and audit (what was queried) complement each other.