Which Python Packages Work in an Online Compiler? 24 Tested

Our online Python compiler runs Pyodide — a full CPython (3.13 on the default build, Pyodide 0.28.0) compiled to WebAssembly that runs entirely in your browser tab. There's no server and no pip. When you hit Run, it scans your imports and loads packages two ways: Pyodide's own pre-built WASM wheels for the big scientific libraries (NumPy, pandas, SciPy…), and `micropip` to fetch pure-Python wheels for everything else. A package works if a WebAssembly-compatible wheel exists; it fails if the library needs a C extension no one has compiled for WASM.

Each package was installed and then imported on the Pyodide runtime the compiler uses. "Load" is the first-time install/import time; note that shared dependencies mean some packages are near-instant once another package has already pulled them in (Pillow arrives with Matplotlib, python-dateutil with pandas), and in the browser you also pay Pyodide's one-time startup (~2–4s).

20 of 24 worked. The whole common data/analysis stack is there, plus the everyday utilities. The four failures split into three distinct reasons — worth understanding, because they tell you what will and won't work for packages I didn't test.

Everything you'd reach for in a notebook works: NumPy, pandas, SciPy, scikit-learn, Matplotlib, SymPy, NetworkX, Pillow and even OpenCV — these ship as Pyodide's own WebAssembly wheels (you'll see them download as ...-pyodide_2025_0_wasm32.whl), so they're fast and reliable. On the utility side, requests, BeautifulSoup, lxml, PyYAML, Pydantic, Jinja2, Click, Rich, regex, python-dateutil and cryptography all install and import cleanly via micropip. If a package is pure-Python or is in Pyodide's built distribution, it just works.

Packages that rely on a C/C++ extension nobody has compiled for WebAssembly can't be installed. In the test that was psycopg2 (needs the native libpq Postgres client), PyTorch and TensorFlow — micropip refuses them with a clear message:

ValueError: Can't find a pure Python 3 wheel for 'psycopg2'.
ValueError: Can't find a pure Python 3 wheel for 'torch'.
ValueError: Can't find a pure Python 3 wheel for 'tensorflow'.

That message is the tell: "no pure Python 3 wheel" means the library needs compiled code that isn't available for WASM. Heavy ML frameworks (PyTorch, TensorFlow), database drivers (psycopg2, mysqlclient) and anything wrapping a system library generally fall here. The rule of thumb: if pip install normally builds C code on your machine, it probably won't run in the browser.

FastAPI is pure Python and *should* work — but it failed, and not for the reason you'd expect:

ValueError: Requested 'typing-extensions>=4.15.0', but
typing-extensions==4.14.0 is already installed

FastAPI pinned a newer typing-extensions than the version Pyodide had already loaded, so the install aborted. This is a common and easily-misread failure: the package isn't unsupported, a transitive dependency just clashed with what's already in the environment. The fix is usually to pin a compatible version yourself (e.g. install an older FastAPI, or the specific typing-extensions first) — worth knowing before you conclude a pure-Python package "doesn't work in the browser."

requests installs and imports fine — but a browser tab has no raw network sockets, so requests.get(...), urllib and the socket module can't make real connections. This is a WebAssembly/browser sandbox constraint, not a bug in the compiler, and it applies to every browser-based Python. For HTTP inside Pyodide you use its own fetch bridge instead:

from pyodide.http import pyfetch

resp = await pyfetch("https://example.com/data.json")
data = await resp.json()

You usually don't have to do anything: just import numpy and hit Run — the compiler detects the import and installs the package automatically. For a specific version or a package whose import name differs from its PyPI name, add a requirements.txt (e.g. pydantic==2.10.6) and it's installed via micropip before your code runs. No pip, no terminal, nothing to set up locally.

Run any of these packages yourself in the browser — no install, no signup. Our online Python compiler loads the package on Run and executes real CPython 3.13 (with 3.12 and 3.11 available too). Paste a NumPy or pandas snippet and watch it install and run in the tab.