SuperAPI AI Directory

workers-fetch

Deployment & operations

View repository

Updated 10-16 05:42

About this skill

workers-fetch

workers-fetch is a tiny curl-like CLI that uses Cloudflare's unstable_startWorker() to spin up a Worker from its wrangler.json(c) / wrangler.toml, send a single request, and print the response. Use it when you would otherwise start bun run dev in one terminal just to issue one curl in another — workers-fetch does both in one shot, then exits.

Install

npm install -g workers-fetch

Usage

# GET / — auto-detects wrangler.jsonc in cwd
workers-fetch

# With a path
workers-fetch /skills/hono
Other methods, headers, body

workers-fetch -X POST -H 'content-type: application/json' -d '{"a":1}' /api/foo

Pin a config file

workers-fetch -c wrangler.toml /api/test

Longer timeout (default 3s)

workers-fetch --timeout 10 /api/slow


Output is JSON with `status`, `statusText`, `headers`, and `body` — pipe through `jq` to extract a field:

```sh
workers-fetch /skills.json | jq '.body | fromjson | .skills[].id'

When to reach for it

  • One-off response checks during development. No long-running dev server.
  • Quick checks of a route or middleware change without context-switching terminals.
  • Scripted local smoke tests in a package.json script or CI step.

When bun run dev is still better

  • Iterating with live reload — workers-fetch re-boots the Worker on every invocation.
  • Hitting the Worker repeatedly from a browser or another tool that expects a real listening port.

Related

  • rj — same idea, but for any HTTP endpoint over the network (not just local Workers).
  • cloudflare — the runtime this tool drives.