At Microsoft, our focus is on offering the very best expertise for Python builders – each for on a regular basis coding and for the agentic workflows which might be more and more central to how builders construct software program.
An amazing sort checking expertise is a core a part of that, not just for surfacing correct and actionable diagnostics and offering a robust basis for Intellisense, however because the bedrock of static semantic intelligence for software program brokers.
Immediately we’re asserting the Kind Server Protocol (TSP) and our partnership with Meta to deliver Pyrefly help to Pylance.
Introducing the Kind Server Protocol
TSP is an open customary that extends LSP – the Language Server Protocol that underpins editor integrations throughout the business – designed to permit any editor or device to combine sort checkers as first-class backends.
Till now, Pylance and Pyright had been tightly coupled. TSP creates a clear separation between the 2: Pylance turns into the VS Code IDE intelligence layer for Python, and the kind checker turns into an interchangeable static evaluation element beneath it.
Pyright stays the default and continues to energy the expertise for the overwhelming majority of customers.
However for builders who’ve a robust desire – or whose codebase is healthier served by a unique checker – switching is now a simple operation, proper from the VS Code standing bar.
What stays fixed no matter which checker you employ is Pylance’s completions, Fast Information, atmosphere administration, and critically its deep integration with GitHub Copilot.
These experiences are constructed on Pylance, not on any particular person sort checker.
TSP means you’ll be able to deliver the checker you need with out giving up any of these options that and love.
Partnering with Meta to deliver Pyrefly to Pylance
We’re excited to announce that we’ve partnered with Meta to design and implement the TSP protocol and guarantee that Pyrefly absolutely helps it.
This collaboration is constructed on a shared perception that the Python ecosystem is healthier served by interoperability than by fragmentation.
Meta brings deep experience in quick static sort evaluation and a selected power in AI/ML library help.
Microsoft brings the VS Code IDE integration layer and the GitHub Copilot expertise.
TSP is the protocol that makes them work collectively.
Meta has been on the forefront of Python typing longer than nearly anybody.
Since 2017, they’ve been fixing sort checking at scale on a few of the world’s largest Python codebases, together with Instagram.
That have formed each design resolution in Pyrefly, which checks code at 1.8 million strains per second and was constructed from the bottom as much as ship a constant expertise in each CI and the IDE.
Meta has additionally invested closely within the broader ecosystem, bringing NumPy to just about 90% sort completeness, finishing pandas‘ public API sort protection, and publishing essentially the most complete public efficiency benchmarks for Python sort checkers.
We’re proud to accomplice with Meta on bringing Pyrefly to Pylance and we stay up for the ecosystem rising round this open customary, and to bringing much more option to Python builders sooner or later.
Why Pylance + TSP + Pyrefly as an alternative of Pyrefly standalone?
That is the correct query to ask.
Pyrefly already has a extremely performant VS Code extension and is a wonderful selection for sort checking in steady integration (CI).
Why use it by means of Pylance as an alternative of immediately?
Pylance gives tight integrations with the GitHub Copilot expertise {that a} standalone sort checker extension can’t replicate – context provision for AI completions, clever code understanding that feeds into GitHub Copilot Chat, MCP instruments for Python codebases, and the total suite of Python developer tooling that Pylance has constructed over years in manufacturing.
In decoupling Pyright from Pylance we now have a transparent separation of issues the place pylance can present an intelligence layer for Python builders and delegate the heavy lifting of sort checking.
By utilizing Pyrefly with Pylance by way of TSP, you get Pyrefly’s sort evaluation powering the engine, with Pylance’s AI-aware intelligence layer on high. You don’t have to decide on.
How you can strive it
TSP help with Pyrefly requires a Pylance model of 2026.4.1 or newer.
Right here’s how the expertise works.
Computerized detection
While you open a workspace, Pylance checks your Python environments for a Pyrefly set up.
If it finds one – whether or not put in in your digital atmosphere or signaled by a pyrefly.toml or device.pyrefly config in your pyproject.toml – you’ll see a notification:
Pyrefly was present in your Python atmosphere. Would you want to make use of it as your supply of diagnostics in VS Code?
Click on Swap to Pyrefly and Pylance will activate Pyrefly as your diagnostics supply instantly, registering the trail to the Pyrefly executable robotically.
No guide configuration wanted.
and the selection to “Swap to Pyrefly” can all the time be reverted or reenabled by way of Settings.
Handbook setup by way of settings
You too can swap at any time by means of the VS Code Settings UI.
Below Python › Evaluation: Diagnostics Supply, you’ll discover a dropdown with three choices:
- Pylance – default, makes use of Pylance’s built-in evaluation
- Pylance + Pyright – explicitly routes diagnostics by means of Pyright
- Pylance + Pyrefly – makes use of Pyrefly because the diagnostics engine
When Pylance + Pyrefly is chosen, Pylance will seek for a Pyrefly set up in your digital environments or obtain it robotically.
A companion Python › Evaluation: Pyrefly Model setting enables you to pin a particular model, level to a neighborhood executable, or depart it empty to all the time use the most recent.
The minimal supported model is 0.60.0.
Standing bar
As soon as lively, the standing bar exhibits Diagnostics Supply: Pylance + Pyrefly alongside your interpreter and atmosphere.
Clicking it opens the Python › Evaluation: Diagnostics Supply setting.
What do you assume?
Attempt the brand new Pylance + Pyrefly expertise immediately in VS Code and tell us your ideas. Share any suggestions by opening a difficulty or voting on an present one. You too can depart a remark beneath or attain us on X.
The Pyrefly group have their personal weblog submit on TSP and the work we’ve achieved collectively.
