Identity-AIops is a governed AI operations tool for self-hosted identity providers (Keycloak + authentik), targeting small and medium-sized infrastructure teams that run their own Keycloak or authentik and want to drive identity operations with agents while retaining full credentials.
In terms of main capabilities, it provides 29 MCP tools: 21 read tools (including 4 analytics), 6 governed write tools, and 2 undo tools (undo_list, undo_apply). The same set of tools supports both Keycloak and authentik, selecting the API shape (authentication flow and resource paths) via each target's platform field. Among these, require_password_reset, rotate_client_secret, client_sessions, client_session_stats, and user_lockout_status are Keycloak-specific shapes and return instructive errors on authentik; authentik, in turn, provides the global session list used for orphaned session checks in stale access audits.
In terms of security and governance mechanisms, every call (including errors) is written to ~/.identity-aiops/audit.db, recording parameters, status, risk level, and optional approver/reason annotations; each session has call and time budgets as well as runaway circuit breakers (IDENTITY_MAX_TOOL_CALLS, IDENTITY_MAX_TOOL_SECONDS, IDENTITY_RUNAWAY_MAX). Risk levels are: reads are low; containment/hygiene writes (disable_user, revoke_user_sessions, require_password_reset) are medium; authorization or boundary-replacement writes (enable_user, update_client_redirect_uris, rotate_client_secret) are high, and this level is a descriptive label rather than a gate. Reversible writes capture and record the real prior state, and record a replayable inverse operation in ~/.identity-aiops/undo.db; irreversible writes record only priorState, and secrets are stored masked. All IdP-returned strings are normalized for injection safety before being presented to the agent. Credentials are stored using Fernet encryption (scrypt-derived key) in ~/.identity-aiops/secrets.enc, and TLS verification is enabled by default.
In terms of usage, it can run as a Claude Code plugin, an OpenClaw plugin, a CLI, or a standalone MCP server. Keycloak requires creating a confidential client with Service accounts roles enabled and granting the required realm-management roles (it is recommended to start with view-users, view-events, view-clients); authentik requires creating an API token for a least-privilege administrator. The CLI covers commands such as users, clients, events, overview, doctor, secret, init, and mcp; write commands support --dry-run and always require double confirmation, and execution is delegated to the governed MCP twin tools, so CLI writes also enter the audit log. When configuring environment variables, note that MCP clients start with a minimal environment and do not inherit shell profile variables; IDENTITY_AIOPS_MASTER_PASSWORD and others must be written in the env block.
In terms of limitations, its verification status is: modeled on the public Keycloak Admin REST API and authentik API v3, and exercised against simulated responses; there is not yet an end-to-end run record against a real instance. The tools do not decide whether a write is allowed; there is no read-only switch, policy file, or approval gate to configure; permissions are determined server-side by the roles of the connected account. The platform does not yet cover Authelia, Zitadel, Ory, and others.
