How to monitor MCP tools/list for breaking changes

AI agents do not read your OpenAPI repo — they call live MCP servers. When a maintainer renames a tool, tightens inputSchema, or removes an entry from tools/list, your agent can fail while HTTP checks stay green.

Why CI diff is not enough

In-repo JSON schema tests and MockDrift fixtures catch changes you ship. They do not see upstream MCP catalogs that change on their own schedule. Production agents depend on the live catalog — that is the contract you need to watch.

Baseline the catalog

1. Add an MCP watch for your server URL (HTTP/SSE transport).
2. Let DriftGuard store the first tools/list response as the baseline snapshot.
3. Schedule checks at Pro (15-minute) or Fleet (5-minute) intervals.

See the MCP monitoring feature overview for console views and alert routing. To wire DriftGuard into Cursor during development, follow the connect DriftGuard MCP in Cursor guide.

Classify diffs before paging on-call

Not every tools/list change is breaking. DriftGuard labels removed tools and required-field additions as breaking, description-only edits as info, and suspicious shape changes as warning. Route breaking events to Slack or PagerDuty; keep the rest in the drift timeline.

Compare two saved catalogs locally with the MCP tools compare tool before you add a watch, or explore the MCP agents solution page for fleet rollout patterns.

Next step

Start on one MCP endpoint with a full Pro trial — no credit card. When the catalog drifts, you get a classified diff instead of a user report. Start free trial →