Drift Detection (MCP)
Monitoring for changes in MCP server tool signatures after approval — new tools added, tool schemas modified, endpoints changed. Alerts fire so teams can re-evaluate before the agent uses new tools.
Drift detection for MCP servers continuously monitors approved tool signatures for changes. When a server adds new tools, modifies existing tool schemas, or changes endpoints after initial approval, drift detection fires an alert — so teams can re-evaluate before the agent calls the new tool.
The rug-pull vector
An MCP server that was approved with 3 benign tools (read docs, search, summarize) might silently add a 4th tool: "delete_all_documents." Without drift detection, the agent will discover and use the new tool immediately — and the admin who approved the server may never know.
Detection strategy
WardenAuth fingerprints every MCP server at registration time — recording the full tool catalog, each tool's input/output schema, and available endpoints. Continuous monitoring compares the live server state against the approved fingerprint and alerts on any diff.
Related terms
Graduated levels of trust assigned to AI agents: low trust requires explicit consent for every tool, high trust auto-approves common patterns. Capped by the user's RBAC permissions.
User-facing interface where a human explicitly authorizes an AI agent to access specific tools within defined boundaries. Time-boxed, revocable, and access-aware.
Technique that identifies and monitors AI agents by behavioral signature. Detects identity changes, tool usage pattern shifts, and privilege drift before they become incidents.