Cloudflare Workers Can Now Hand Your Coding Agent a Debug Trail Automatically

What happened

Cloudflare updated its local development tools so a coding agent can debug your app’s backend on its own, without you copying error messages into the chat first.

As of August 4, 2026, wrangler dev (Cloudflare’s command-line tool for running a Worker, its serverless backend code, on your own machine before deploying it) and vite dev with Cloudflare’s Vite plugin automatically record what your app did while it ran locally: every outgoing network request and how long it took, every read or write to Cloudflare’s storage and database services, and the full path a request took from start to finish. This record is called an OpenTelemetry trace, an industry-standard way of logging what a program actually did step by step, so a person or an AI agent can see exactly where something broke instead of guessing from a single error message.

None of this needs setup. You don’t install a tracing tool, flip a setting, or tell your coding agent to check logs. If Cloudflare’s dev server notices it’s being run inside a supported coding agent’s session, it automatically prints a short hint pointing that agent at a new local API, so the agent can pull the trace data itself.

How the agent uses it

That local API is called the Local Explorer. It’s a REST API, a set of local web addresses a program can call to fetch data, that only runs on your own machine while wrangler dev or vite dev is active. The API publishes a machine-readable list of what it can do, so an agent can work out on its own what it’s allowed to ask for, instead of you spelling it out in your prompt. Through the same API, an agent can also inspect the current state of your local KV store (Cloudflare’s key-value data store), R2 (its file storage), D1 (its SQL database), and Durable Objects (small units of persistent state tied to one instance of your code).

Cloudflare’s own example shows the point: an endpoint that creates orders starts returning server errors after a database schema change. Instead of a person pasting the error into a chat, the agent queries the Local Explorer directly, sees that a database write failed because a column is missing, checks the database schema, finds a migration (a saved set of database changes) that was never applied, applies it, and retests. All of it happens on one machine, with nothing deployed.

What you need to do to get it

Update your tooling: run npm install --save-dev wrangler@latest, or npm install --save-dev @cloudflare/vite-plugin@latest if you build with Vite. Cloudflare’s announcement doesn’t say whether this requires a specific paid plan or is still labeled beta, only that it’s live “starting today.” Treat it as available now to anyone on an up-to-date wrangler dev or vite dev setup.

Who should care

This matters if you build on Cloudflare Workers, especially if a coding agent is the one running your dev server and chasing down bugs for you. It doesn’t apply if you deploy elsewhere, like Vercel, Netlify, or Supabase.

One thing worth knowing plainly: everything described here stays on your own machine. The database and storage state your agent inspects through the Local Explorer is your local development copy, not anything live in production, so this doesn’t expose real user data to your agent or anyone else.

What builders should do next

Update wrangler or the Vite plugin, then give your agent a real bug to find instead of one you’ve already diagnosed for it. A concrete test: rename a database column your code still references somewhere, then ask the agent to find and fix the resulting failure without telling it what you changed.

Run that same test twice, once before you update your tooling (where you paste in whatever error message you get) and once after (where the agent queries the Local Explorer on its own). Compare how many back-and-forth messages it takes the agent to land on the real cause, and whether it names the specific failing piece, like “the D1 insert on the orders table,” on its first attempt rather than guessing broadly.

The bigger shift here isn’t really about Cloudflare. It’s about where debugging information should live: not retyped into a chat window, but somewhere an agent can go check for itself. If it works the way Cloudflare describes, the real change is how much less “give your agent context” ends up meaning “copy and paste an error.”

End of article