Turn your API into tools AI agents can call.
One command reads your OpenAPI spec or validation schemas and writes them: typed, checked for safety, ready to test in the local playground.
What every generated tool ships with.
An agent never sees your API. It sees each tool's name, description, and schema, and nothing else. So each one meets every line of the same standard:
/ CODEGEN
NO. 001The tool standard.
every generated tool, checked against all eight on every run
01
A name that says what it does
post-trips-trip-id-story-generate→ generate-story
02
A description an agent can act on
What it does, when to use it, what it returns.
03
Inputs spelled out in words
Required, optional, limits: written in, never guessed.
04
Unreviewed tools never reach agents
Auth, admin, and destructive ones stay unregistered until you turn them on.
05
Failures that say what to try next
A readable reason, never a stack trace.
06
A human keeps the final click
Payments and deletes stop at a confirmation.
07
Re-runs never touch your code
The spec moves, the tools update, your edits survive.
08
No account, no telemetry, no runtime
Plain TypeScript in your repo, nothing imported back.
Point it at your API.
The first run writes nothing. It only shows you the tools hiding in your API.