Short answer: A generated server shows exactly which operation and schema produced each tool, and surfaces the operations it skipped because their metadata was incomplete. A catalog server is opaque: you trust the author and discover weak tools at runtime. Provenance is the difference between a tool you can audit and a black box.
What provenance gives you
Every tool traces to an operation, its operationId, and the request schema. When an agent calls a tool, you can see the exact endpoint behind it. That is the level of confidence you need for a service you own.
Why skipped tools matter more than missing ones
A skipped tool is visible: the studio tells you an operation was not compiled, usually because it lacks an operationId or has ambiguous metadata. You can fix the contract and regenerate. A silently weak tool is worse — it exists, the agent calls it, and it fails because its parameters were guessed.
The contrast with a catalog
| Property | Catalog server | Generated server |
|---|---|---|
| Tool origin | Trust the author | Visible per tool |
| Weak tools | Found at runtime | Surfaced at build |
| Fix | Wait on the author | Edit the spec, regenerate |
The discipline
Review the tool list and the skipped entries before pointing an agent at the server. The skipped tools are a punch list for the contract, not a defect to ignore.
This is part of the Powerduck vs Smithery comparison. Finish with registry vs local-first.
Top comments (0)