The glue code is gone
For two years, running Python on Cloudflare Workers meant accepting a boundary. You wrote Python, but the moment you touched a Cloudflare binding — a Queue, a Durable Object, a KV namespace — you dropped into TypeScript-shaped glue to convert objects across the RPC line. According to the Cloudflare Blog announcement, that conversion is now handled inside the Workers runtime and the Python SDK. Python Workers reached general availability on September 21, 2026, and the practical effect is that you can use platform bindings without writing JavaScript.
That matters less as a syntax win and more as a failure-mode win. The announcement is explicit that the old conversion step was “a common source of error for both humans and AI agents.” If you have ever had a coding agent generate a Worker that looked right and failed at the binding boundary, this is the class of bug being removed. The same reasoning shows up in our earlier post on giving coding agents production evidence instead of just the repo — the fewer invisible translation layers between intent and runtime, the less your agent has to guess.
Frameworks run, but the platform is the server
The GA release supports FastAPI, Django, and Flask through workers.asgi and workers.wsgi connectors. The design choice here is worth understanding before you port anything.
In a normal deployment, Uvicorn or Gunicorn owns concurrency and connection handling while your framework owns application logic. Cloudflare’s stated position is that Workers itself is the web server, so the connectors act as a thin bridge: they translate the incoming request into standard WSGI/ASGI structures and pipe the response back. You keep your framework’s code organization; you do not get a Python process you can tune. Anything you previously solved with worker counts, thread pools, or server-level middleware needs rethinking.
Sockets, drivers, and the database question
Python Workers previously had no TCP socket support, which meant database drivers like asyncpg and aiomysql simply could not connect. Cloudflare implemented socket system calls on top of the Workers connect API, translating Python socket operations into the JavaScript calls the runtime uses. That bridge is what makes Hyperdrive integration possible, and it is also what lets HTTP clients like requests and httpx work.
The knock-on effect is the AI tooling story. The announcement states that openai, langchain, and mcp can now run natively in Python Workers, combinable with Workers AI for inference or AI Gateway for proxying. If you are building agent pipelines, that removes a common reason to keep a separate container running.
The package ecosystem is the real constraint
Because Python Workers run in a WebAssembly sandbox, any package with native C, C++, or Rust extensions must be cross-compiled. Cloudflare previously compiled and hosted these packages manually, which capped what you could use. The fix was standards work: PEP 783, accepted after more than a year of discussion, defines the PyEmscripten platform, and the Pyodide toolchain plus cibuildwheel support were extended to build for it.
Read that carefully before you plan a migration. The announcement says the ecosystem is still adopting the standard and that Cloudflare is working with maintainers to add builds. The supplied material does not list which packages are ready today — it points to the Hyperdrive documentation for currently supported packages. So the honest position is: check your dependency tree first, and treat unsupported native packages as a real blocker rather than a temporary annoyance.
What to do with this
The lowest-risk first move is a Worker that touches one binding and one external HTTP call — no native dependencies. That exercises the two things GA actually changed: Pythonic bindings and working networking. If that holds, the framework connectors and Hyperdrive become the next question, and package availability becomes the gate on everything after.
GA here means supported and production-ready, not finished. Cloudflare says performance, memory efficiency, and broader package support are still on the roadmap.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.
