Drift Detection

Explain continuous monitoring and how to respond to drift.

What is Drift?

Definition: Any difference between the locked state and the live server state.

Causes:

  • Server updated (new version deployed)
  • Configuration change (different flags)
  • Supply chain attack (malicious modification)

Detecting Drift

Command:

mcptrust check -- "npx -y @scope/server /args"
# Exit 0 = no drift
# Exit 1 = drift detected
# Exit 2 = runtime/usage error

Detailed Output:

mcptrust diff -- "npx -y @scope/server /args"

Shows: tool name, drift type, severity, before/after values.

Drift in CI

Example GitHub Action:

- uses: mcptrust/mcptrust/.github/actions/mcptrust@v0.1.1
  with:
    mode: check
    server_command: 'npx -y @scope/server /tmp'
# Fails PR if drift detected

Drift at Runtime (Proxy)

Command:

mcptrust proxy --lock mcp-lock.json -- "npx -y @scope/server /tmp"

Behavior:

  1. Before bridging traffic, performs preflight drift check.
  2. If drift exceeds --fail-on threshold, refuses to start.

Responding to Drift

Decision Matrix:

ScenarioAction
You deployed a new server versionRe-lock, review changes, re-sign
Upstream released a patchReview release notes, re-lock if safe
Unexpected new tool appearedINVESTIGATE. Possible attack.
Description typo fixedRe-lock (low risk)
Schema changedCarefully review new arguments. Re-lock only if safe.
Docs — MCPTrust