What is contract drift (and why uptime checks miss it)

Uptime monitoring answers one question: did the endpoint respond? Contract drift answers another: did the shape of the response still match what your code expects? Vendors change JSON schemas silently — required fields disappear, enums gain values, MCP tools vanish from tools/list — while returning HTTP 200.

Contract drift vs. downtime

• Downtime: connection failures, 5xx, timeouts — traditional monitors catch these.
• Contract drift: schema changes that break parsers, agents, and integrations without tripping uptime.

Agent teams feel this first: a removed MCP tool fails mid-workflow even though the server is “up.” Platform teams see it as deserialization errors in production while synthetic checks stay green.

Why fixtures and CI are not enough

Checked-in OpenAPI specs and mocked responses test what you ship. They do not observe third-party APIs that change on their own cadence. Pair in-repo diff with live vendor monitoring so merge-time gates and scheduled watches cover both sides of the boundary.

How DriftGuard classifies change

Baseline a live URL or MCP catalog, diff on a schedule, and label each change breaking, warning, or info. Route breaking events to Slack or PagerDuty; keep the rest in an audit timeline for postmortems.

Explore drift detection, API monitoring, and MCP monitoring — or paste two JSON blobs into the compare tool to see classified diffs instantly.

Start on one endpoint

Start a free trial on a single vendor URL or MCP server. When contract drift lands, you get a diff — not a vague “something broke in prod.”