Just talk to your Cloudflare dashboard: meet Agent Lee!

Hey everyone, it's Shii-chan! Today I found an announcement that flips the whole idea of a dashboard on its head. Ready to stop bouncing between tabs?

Cloudflare Blog blog.cloudflare.com

What was announced?

On the Cloudflare Blog, Agent Lee just launched in beta — an AI assistant built right into the dashboard. It understands what lives in your Cloudflare account, and you drive it by talking in plain language instead of clicking around.

The interface of technical products hasn't really changed since the early days of the Internet: clicking five pages deep, cross-referencing logs across tabs, hunting for hidden toggles. Agent Lee's bet is to replace all of that with a single prompt.

Agent Lee isn't answering FAQ questions — it's doing real work, against real accounts, at scale.

It's already served over 18,000 daily users in active beta, executing nearly 250k tool calls per day. That's a lot of real work!

Why it matters

Picture your Worker returning 503s at 02:00 UTC. Is it an R2 bucket, a misconfigured route, or a hidden rate limit? You end up opening half a dozen tabs, hoping to spot the pattern. Most of us don't have a teammate who knows the whole platform standing over our shoulder at 2 a.m.

Agent Lee wants to be exactly that teammate — and it won't just find the cause, it will fix the problem on the spot. That's the real difference from a FAQ bot.

What changes

With natural language, you can:

• Ask about your account: "Show me the top 5 error messages on my Worker."
• Debug an issue: "I can't access my site with the www prefix."
• Apply a change: "Enable Access for my domain."
• Deploy a resource: "Create a new R2 bucket for my photos and connect it to my Worker."

Instead of switching between products, you describe what you want and Agent Lee gets you there with instructions and visualizations. Its Generative UI is the fun part: ask about your error rate over the last 24 hours and it renders a chart inline from your actual traffic, instead of sending you off to a separate Analytics page.

Every conversation sits inside an adaptive grid. You drag across it to carve out space for new UI blocks, then describe what you want to see. Dynamic tables, interactive charts, architecture maps and more turn your chat history into a living dashboard. It covers DNS, Workers, SSL/TLS, R2, Registrar, Cache, Cloudflare Tunnel, API Shield, and more.

Dive Deep

The heart of Agent Lee is Codemode. Rather than showing MCP tool definitions to the model directly, it converts the tools into a TypeScript API and asks the model to write code that calls it.

This works better for two reasons. LLMs have seen a huge amount of real-world TypeScript but very few tool-call examples, so they're more accurate writing code. And for multi-step tasks, the model can chain calls in a single script and return only the final result, skipping the round-trips.

So what stops it from writing to your account on a whim? The generated code is sent to an upstream Cloudflare MCP server for sandboxed execution, but first it passes through a Durable Object acting as a credentialed proxy.

The DO inspects the method and body to classify the code as read or write. Reads are proxied directly; writes are blocked until you explicitly approve them through the elicitation gate. API keys never appear in the generated code — they're held inside the DO and injected server-side. So it's not a throwaway sandbox, it's a permission architecture that structurally prevents writes without your approval.

Cloudflare's own MCP server exposes just two tools: a search tool to query API endpoints, and an execute tool to write code that performs API requests. Writes always go through elicitation, and Agent Lee cannot skip that step. The confirmation prompt you see is the gate, not a UX courtesy.

And it's all built from the same parts you can use: Agents SDK, Workers AI, Durable Objects, and the same MCP infrastructure. No hidden internal tools — just the Cloudflare building blocks you already have. Quality is measured continuously with evals (conversation success rate and accuracy), thumbs up/down feedback, tool-call success rate and hallucination scorers, and a per-product performance breakdown.

Wrap-up

• Cloudflare launched Agent Lee in beta: an AI assistant you talk to in the dashboard to run tasks.
• Questions, debugging, config changes, and resource creation all happen from one prompt, with a Generative UI that draws charts and tables inline.
• Under the hood, Codemode turns MCP tools into TypeScript, and a Durable Object classifies reads vs writes so every write needs your approval.
• Every piece — Agents SDK, Workers AI, Durable Objects — is the same stack any Cloudflare developer can use.
• It's already running at 18,000 users and 250k tool calls a day. It's in beta for Free plan users: log in to the Cloudflare dashboard and click "Ask AI" in the upper right corner to start.

This one lands for Cloudflare users tired of 2 a.m. troubleshooting, and for developers who want to build AI agents on their own platform. Got feedback or a feature wish? You can share it here.