Cloudflare launched cf, a CLI that gives coding agents the whole Cloudflare API instead of the 280 commands in Wrangler
Cloudflare launched cf on 2026-09-28, a new command-line tool that exposes the whole 3,000-operation Cloudflare API to a coding agent, instead of the 280 operations Wrangler covers, with JSON output as default and a TypeScript config file.

Image: Cloudflare
Why it mattersA coding agent asked to buy a domain, put a Worker behind it and add a WAF rule used to need three tools and a browser tab. cf reduces that to one command tree and one config file the agent can type-check.
Almost half of the commands run through Cloudflare's Wrangler CLI last week came from a coding agent, not a person. Cloudflare's blog post on 2026-09-28 puts the number at 48%, up from about a quarter in March 2026 and single-digit percentages a year before that. The same post launched cf, a new command-line tool designed so that share can keep growing.
Cloudflare says Wrangler exposes 280 operations, most of them centred on Workers, while cf covers "the entirety of the Cloudflare API surface of over 3,000 operations". The company gives the example of one agent workflow: "set up a worker, deploy it, monitor and observe it, protect it with Cloudflare Access, buy a domain, and front it with Cloudflare WAF, all from a single tool", where each of those used to reach a different console, API endpoint or third-party CLI.
Built for the reader who cannot see
Every design choice in cf points at that agent-heavy reader. JSON is the default output, "pretty printed for humans and condensed for agents for maximum context savings", so an agent that wants to reason about a response does not have to pass --json on every call. A new cf cli search command lets an agent describe the task it is trying to do in natural language and receive a list of commands whose API descriptions and parameters match, which is faster than making the model page through a --help tree.
Configuration also becomes something an agent can type-check. cf reads a cloudflare.config.ts file at the project root, so an editor or language server sees the whole shape of a Workers deployment, DNS zones, WAF rules, Access policies and more, and can catch invalid values before an agent runs anything. Cloudflare writes that "typed configuration is enormously helpful for agents", and that language-server hints turn into more accurate suggestions when a model asks what to write next.
Forge, the generator behind it
The reason cf can cover 3,000 operations on day one is that Cloudflare did not hand-write it. cf is generated from Cloudflare's OpenAPI schemas by an internal tool called Forge, which the company open-sourced under Apache-2.0 the same day. Forge produces the CLI, the JavaScript SDK and the documentation from one source, which is how each SDK stays in step with the API. Cloudflare says Forge replaces its previous mix of code generators, including the Stainless-generated SDK it had used until acquiring Stainless earlier this month.
Cloudflare has posted the cloudflare/cf repository on GitHub under Apache-2.0. The cf migrate command converts an existing Vite-based Wrangler project to cloudflare.config.ts, and cf continues delegating to Wrangler for esbuild projects during the transition. Cloudflare says a final major Wrangler version will point users and agents at cf when the beta ends, followed by an 18-month maintenance window for Wrangler.
The bet
For teams whose Cloudflare work already flows through an agent, cf is an easier target than the API plus Wrangler plus a browser tab. It removes the reason to teach a model three different ways to reach the same service. The question left open is whether the whole 3,000-operation surface is safe to hand to an agent without another layer between them, or whether that job stays with pre-commit review of the config file and whichever guardrail a shop already has on agent tool use.
Source
- Cloudflare: Introducing cf: the agentic CLI for the entire Cloudflare API, 2026-09-28.
- GitHub: cloudflare/cf, Apache-2.0.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.

