Detect MCP Tool-Call Drift After an Upgrade | Reelier
Evidence state: explained against released reelier@0.32.0 on 2026-08-12.
Record a fixed MCP or HTTP workflow before the upgrade, replay it against the upgraded dependency or server, and use diff to compare step structure and recorded outcomes. This can expose covered drift, but it is not a comparison of arbitrary raw response data and does not evaluate model reasoning, planning, or tool selection.
Evidence note
The released README documents replay and diff behavior. This page is explained rather than reproduced because the cluster retains no before-and-after upgrade fixture.
Fix the workflow before upgrading
Record the calls and declared checks while the dependency or server is at the accepted version. That artifact defines the behavior the later replay will cover.
Replay after the change
Run the fixed workflow against the upgraded target. Failed declared checks expose covered outcome changes at replay time.
Read the diff narrowly
diff compares recorded step structure and outcomes between runs. It is not a general raw-response diff and it says nothing about native shell or file actions, unrecorded branches, or model-selected calls outside the fixed workflow.
Non-claims
This does not prove that an action was safe or semantically correct.
This does not prove that every write was observed or receipted.
Falsifier
A changed covered step structure or recorded outcome that produces neither a failed declared check nor a run diff would falsify this claim.
Citation-friendly definition
Replaying a fixed recorded workflow after an upgrade can expose covered changes in step structure and recorded outcomes; it does not compare arbitrary raw responses or model reasoning.