The five tools

Orstrum registers five MCP tools. Every result is plain text of at most ten lines, because every token of output comes out of the limit Orstrum is protecting. The agent is told when to call each one; you rarely need to.

orstrum_status · all tiers

One line: weekly, 5-hour and binding-limit usage, the policy mode, context size, and loop state. The agent calls it before a large task or when unsure how much budget is left.

Orstrum: week 3% · Fable 5%, resets Oct 17 8pm · 5h 4% · measured · mode normal · context 27k · loops not tracked yet

measured means the numbers came from the usage endpoint; estimated means the token-count fallback. Readings older than two hours show as budget unknown. The per-model limit appears only when the session's model has one tighter than the all-models limit.

orstrum_checkpoint · Pro

Snapshots the project to the shadow store. Takes an optional label. Called after a test, build or run succeeds and before a risky change. See Checkpoints and rollback.

orstrum_rollback · Pro

Restores the last working checkpoint, or a named one (checkpoint: "cp-4"). Takes a safety checkpoint first and names it in the result so the rollback can be undone. Called when repeated fixes keep failing or when you want to undo.

orstrum_handoff · Pro

Stores the agent's summary of the session, validated against the event log. Called when context is bloated or the limit is near. See Handoff and recall.

orstrum_recall · Pro

Returns the project's spec, rules and last handoff, at most ten lines. Called at session start or when unsure of earlier decisions. The same text is injected automatically after compaction, resume or fork.

On Free

The four Pro tools are registered with their real descriptions and answer with one line: orstrum_rollback is available on Orstrum Pro. The agent sees the same triggers on every tier, so upgrading changes what the tools do, not how the agent behaves.