Understand a change in your MCP tool inventory.
See what a scheduled connection observed before deciding whether a tool needs its own check.
Read the observations in order
- Save an authorized public HTTPS endpoint and configure Endpoint access if it requires credentials.
- Inspect the first inventory observation, then compare a later scheduled observation for added, removed or changed tools and schemas.
- Review connection errors and gaps separately. An inventory change does not establish that a tool was called or that its output is correct.
Illustrative example
For the reserved endpoint https://mcp.example/mcp, suppose two observations contain:
| Observation | Tool names | Evidence |
|---|---|---|
| Earlier | search_docs | One listed tool |
| Later | search_docs, lookup_version | An added tool name |
This example shows an inventory difference, not an executed tool, a tested integration or a customer result.
Keep timing and availability separate
Inventory checks exchange initialize, notifications/initialized and tools/list; they do not send tools/call. Recorded inventory time covers the connection and protocol exchange, not separately measured first-token or internal model time. An on-demand setup test also includes its own queue/authentication duration and does not add scheduled availability history.
Availability is successful observed checks divided by observed checks. Scheduled samples can miss short outages. Gaps remain unknown; pause and maintenance are not successful observations. Publishing numeric availability requires a separate owner action and does not publish private inventory. Current cadence and retention are shown on the pricing page.
Worked monitoring examples
How MCP connection monitoring works → · Understand tool checks →