Create reusable Toolchains

1. Describe what should be reusable

1. In Desktop, open Setup (MCP, etc.) → Toolchains and choose the capability generation action.
2. Describe the capability and which inputs vary. Example: summarize a set of documents into a table; documents, output language, and length vary on every call.
3. Review the generated purpose, input/output contract, examples, and implementation. Keep project-specific filenames out of a general-purpose capability.

2. Validate the saved version

Open the capability and inspect Validation. Run its examples before making it callable. For examples that write files or reach an external service, review the effects and explicitly permit that validation. A failed version remains unavailable for normal calls.

3. Call and inspect

1. Choose a saved version, enter input JSON matching its contract, and select Call this version.
2. Inspect the receipt and Call history, then open the actual output. View implementation graph lets you inspect how it runs.
3. Inspect whether the capability is eligible for discovery; not every generated item is callable or searchable.

4. Improve with a new version

Use the improvement action to describe the change. The new version stays in the same catalog item and preserves earlier saved versions. Review and validate it before using it. Calling a Toolchain does not turn on its implementation graph's schedule.