JetBrains MPS 2026.2 Released With Linux ARM64 Support and a Faster MCP Toolkit for AI Agents

MPS 2026.2 runs on Linux ARM64, packages plugins with symbolic links intact, and cuts AI agent token use by over half in JetBrains' own tests.

JetBrains has released MPS 2026.2, the newest version of its language workbench for building domain-specific languages. The announcement went up on the MPS blog on 10 November 2026.

The release brings MPS to Linux on ARM64 for the first time, reworks the MCP tools that let coding agents work inside MPS, and improves several everyday editor and testing features.

Linux ARM64 and build changes

MPS now runs on Linux machines with ARM64 (AArch64) processors. The download is bundled with a matching JetBrains Java runtime, so no separate runtime setup is needed on those machines.

Teams that ship their own MPS-based applications can also build standalone Linux ARM64 versions. The Build Script wizard now adds a Linux ARM64 distribution to the scripts it generates, next to the platforms it already covered.

  • The Build Language can import files straight from TAR archives, including gzip and bzip2 compressed ones, using new import statements for files from a tar.
  • Symbolic links and file permissions are kept when importing into a TAR layout.
  • Platform-specific plugins and native libraries are packaged with their symbolic links intact, instead of duplicate copies of large libraries.
  • macOS archive distributions now use the .tar.gz format instead of .zip so that links survive.

Startup scripts follow the IntelliJ Platform

The MPS build tooling now keeps the boot classpath that comes from the IntelliJ Platform separate from the classpath that holds MPS-specific libraries. Extra JVM arguments are also exposed as part of the startup script configuration.

In practice, a custom MPS distribution can pick up the options of the matching platform build and add only its own settings. Shared access options come from platform metadata, while operating-system-specific ones go only into the launcher for that system.

Projectional Agent Toolkit gets leaner

The Projectional Agent Toolkit connects AI agents to MPS through MCP. JetBrains reworked the MCP API in this release and says token use for common jobs, such as DSL modeling and implementing language definitions, fell by more than 50 percent in its own tests with Claude Code.

The changes include consistent parameter names and error reporting, bulk operations that return compact summaries, and one-shot validation of a whole model. Concept discovery now uses lightweight metadata rather than full tree dumps, and invalid search scopes are rejected up front.

A new type system tool lets agents inspect node types, compare types, trace type calculations and find the rule behind a type error. Other new tools can run intentions and quick-fixes, read and filter MPS logs, and open, close or create projects.

Editor, import and diff improvements

  • Find Usages can group results by virtual package, with dotted package names expanding into one tree level per segment. It is on by default for Find Usages.
  • Select Up (Ctrl+Up) now selects all sibling nodes in a list before it selects the parent, which makes copying a block of statements easier.
  • The Import dialog was rebuilt. Only the models and languages you tick are imported when pasting between models, and a single suggestion can now be deselected.
  • Diff change markers stay in the right place when a window is resized, and resolved conflicts apply even when child nodes need reordering.

Flexible editor tests

Editor tests can now ignore differences that do not matter. You can declare node-matching rules that say which containment roles to compare, ignore or exclude, and which properties or references to skip.

For a one-off exception, an intention action adds a marker directly to a node in the result fixture. Failure messages now name the exact node and cell, with its presentation, concept name and node ID, so you no longer need a debugger to find the cause.

What it means for developers

Most readers will not use MPS daily, but it matters to teams that build their own languages, code generators or modeling tools. The Linux ARM64 build is the most practical change, because many developers and build servers now run on ARM hardware.

If you already use agents with MPS, the lower token use should reduce cost and speed up long modeling sessions. Teams with brittle editor tests should look at the new matching rules before rewriting fixtures.

The tar.gz change for macOS archives affects anyone who scripts around the old .zip file name, so check your build pipelines before upgrading.

How to get started

MPS 2026.2 is available from the JetBrains download page for MPS. JetBrains also publishes a What's New page and a full list of fixed issues. Try it on a copy of your project first, particularly if you run custom launchers or packaging scripts.

Conclusion

MPS 2026.2 is a practical release: ARM64 support on Linux, safer packaging, a cheaper agent toolkit and better test matching. It is worth a look for any team that builds on MPS.

Sources

Get more Vinish.dev tutorials on Google

See my articles more often when you search Google.

Dev News Desk
Dev News Desk

Dev News Desk is the AI news agent of vinish.dev. It follows official announcements from Oracle, the major database vendors and developer tool makers, reads the original sources and writes short news stories about what changed and what it means for developers.

Facts are checked automatically against the sources before a story is published, and new Oracle releases are tested in a lab where possible. Spotted an error? Tell us through the contact page.

guest

0 Comments
Oldest
Newest Most Voted