OpenSumi core turns a clone into an IDE with four yarn commands
A framework helps you quickly build AI Native IDE products. MCP Client, supports Model Context Protocol (MCP) tools via MCP server.
At a glance
- What is it?
- opensumi/core is a TypeScript monorepo for building IDE products, published as @opensumi packages and presented as an AI native IDE framework with MCP client support. Running it locally takes yarn install, yarn run init and yarn run start, while the real decisions are which of the six example projects you start from and which runtime you target.
- Who is it for?
- OpenSumi core suits a team that needs an IDE shell rather than a text editor with plugins, and that will pick one of the six example projects as its starting point instead of assembling a shell from parts. It does not suit someone who only wants a browser IDE to run, since CodeBlitz and the lite IDE are separate repositories.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 31 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- opensumi/core3,666 ★ · +9
| Repository | 2026-09-11 | 2026-11-15 |
|---|---|---|
| opensumi/core | 3.7k | 3.7k |
9 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Four commands take you from a clone to a running IDE
The development path is short enough to copy whole:
$ yarn install
$ yarn run init
$ yarn run download-extension # Optional
$ yarn run startThe third line is marked optional, so a first run can skip the extension download and add it later. By default the start command opens the tools/workspace folder inside the project, and pointing it somewhere else takes an environment variable rather than a flag:
$ MY_WORKSPACE={local_path} yarn run startThat variable is the first thing to know about the local setup, because the default workspace is a sample folder rather than your own code. The README also warns that system level environment dependencies remain, and sends you to the development environment preparation section of CONTRIBUTING.md for those, which means the four commands are the JavaScript side and not the whole prerequisite list.
Six example projects decide which IDE you are actually building
The Getting Started section is really a menu of shells, and the differences are not cosmetic. Cloud IDE points at opensumi/ide-startup, and Desktop IDE based on the Electron points at opensumi/ide-electron for a local application. CodeBlitz is described as a pure web IDE Framework and Lite Web IDE as a pure web IDE on the browser, and both live in separate repositories rather than in this one. CodeFuse IDE is an AI IDE based on OpenSumi, in the codefuse-ai organisation, and The Mini-App liked IDE points at opensumi/app-desktop. So the choice is web versus desktop versus mobile shaped, and the framework repository itself is the dependency rather than the product you ship. Each example carries its own build and packaging decisions, which is why the README offers the list instead of a single quick start.
The build is a fixed sequence of seven sub-builds
The root manifest is named @opensumi/core and marked private, so it is never published as such; what reaches npm are the packages under it. The workspace covers four globs: packages/* plus tools/dev-tool, tools/playwright and tools/cli-engine. The build entry point is compile, which sets NODE_ENV=production and runs the build script through tsx, and build is a thin wrapper that calls it. The full build is build:all, and its order is meaningful because each step produces something the next consumes:
- the prebuilt webview bundle from the @opensumi/ide-webview workspace - the main compile - the worker host and the extension host, both built inside packages/extension - the watcher host from packages/file-service - the components package, which has separate dist and library targets - the Monaco worker from packages/monaco
Building one of these in isolation is a supported move, which is what the individual build:* scripts are for.
Two runtime targets are chosen with add:browser and add:node
The target environment is not fixed, and the manifest has a pair of scripts for changing it: add:browser and add:node, each a tsx call into the scripts directory. That pair explains the shape of the rest of the build, because several bundles exist only for one runtime, such as the prebuilt webview bundle and the Monaco worker. Package boundaries follow the same logic. packages/extension owns both the extension host and the worker host builds, packages/file-service owns the watcher host, packages/components exposes build:dist and a library build, and packages/monaco exposes a worker build. Lerna is present in the root as lerna.json and handles the clean step across those packages, which deletes each lib directory before running the package level clean scripts.
Continuous integration is two workflows and one coverage upload
Two workflow badges sit at the top of the file, one for ci.yml and one for e2e.yml, and a third badge points at codecov for coverage on the main branch. The coverage number is not decoration: the root script ci is defined as yarn run init followed by yarn run test:cov, so the same path runs locally and in the pipeline. Jest is the runner, configured in jest.config.js with three separate setup files, jest.setup.base.js, jest.setup.jsdom.js and jest.setup.node.js, which is the shape you expect when the same test suite has to run against a DOM and against plain Node. There is also a check:dep script that runs a dependency check from the scripts directory, and a changelog script with a changelog-old predecessor kept alongside it.
Deployment descriptors for two hosts sit next to the code
Two files at the root are aimed at hosting platforms rather than at developers: nixpacks.toml and railway.toml. They describe how the project is built and served when a platform builds it from the repository, which is the deployment path for anyone who does not want to assemble their own container. The rest of the root is tool configuration and content: configs, scripts, typings and test directories, tsconfig.json for the TypeScript project, .vscode for editor settings, .husky for git hooks with commitlint.config.js beside it, .prettierrc and .prettierignore plus .eslintrc.js and .eslintignore for formatting and linting, codecov.yml, and a Yarn Berry setup with .yarn and .yarnrc.yml alongside yarn.lock. Two languages of documentation ship together, README-zh_CN.md and CONTRIBUTING-zh_CN.md.
MCP support is advertised in the description and in a badge
The project description names two things: a framework for building AI Native IDE products, and MCP client support for Model Context Protocol tools via an MCP server. The README itself carries two badges for that, one marking the project as a client and one marking the tools feature, both pointing at the Model Context Protocol site. Beyond those badges and the description, the README does not set out how the MCP client is configured or which transport it uses, so the feature is discoverable but not documented in the file itself. The full documentation lives at opensumi.com, and the release notes and breaking changes are kept in CHANGELOG.md, which the README links as the place to check before upgrading.
The newest tag is from 2025 and the branch moved in September
Version history and current activity diverge here. The recent releases are v3.8.1 published on 2025-03-07, v3.8.2 on 2025-03-11, and v3.9.0 on 2025-05-20. The last push to the default branch main is dated 2026-09-29, and the repository is not archived, so work has continued on main for more than a year after that tag without a further release being cut. Anyone pinning a dependency to a tag is therefore pinning code from 2025, while anyone tracking main gets unreleased changes. Licensing is MIT with copyright held by Alibaba Group Holding Limited and Ant Group Co. Ltd., and NOTICE.md covers the third party code that ships under other open source licences.
Editorial conclusion
OpenSumi core suits a team that needs an IDE shell rather than a text editor with plugins, and that will pick one of the six example projects as its starting point instead of assembling a shell from parts. It does not suit someone who only wants a browser IDE to run, since CodeBlitz and the lite IDE are separate repositories. Before building, expect system level dependencies listed in CONTRIBUTING.md, decide whether the optional extension download is wanted, and note that the newest tag is v3.9.0 from 2025 while the branch has moved since.
Frequently asked questions
How do I run OpenSumi core locally?
Run yarn install, yarn run init, then yarn run start, with yarn run download-extension as an optional step before the start. The default workspace is the tools/workspace folder in the project, and MY_WORKSPACE={local_path} yarn run start points it at your own directory instead.
Which OpenSumi example project should I start from?
Six are offered: Cloud IDE at opensumi/ide-startup, Desktop IDE based on the Electron at opensumi/ide-electron, CodeBlitz as a pure web IDE Framework, Lite Web IDE on the browser, CodeFuse IDE based on OpenSumi, and The Mini-App liked IDE at opensumi/app-desktop. They live in separate repositories, so the choice is between web, desktop and mobile shells.
Does OpenSumi core support MCP?
The project describes itself as an MCP client that supports Model Context Protocol tools via an MCP server, and the README carries badges for the client status and the tools feature. The README does not document the configuration, which is left to the full documentation at opensumi.com.
What does build:all do in OpenSumi core?
It runs a fixed sequence: the prebuilt webview bundle from the @opensumi/ide-webview workspace, the main compile, the worker host and extension host from packages/extension, the watcher host from packages/file-service, the components package, and the Monaco worker from packages/monaco. Each step also exists as its own build:* script.
Where are OpenSumi breaking changes recorded?
In CHANGELOG.md, which the README links as the place holding all release notes and breaking changes. The recent tags are v3.8.1, v3.8.2 and v3.9.0, with v3.9.0 published on 2025-05-20 and the last push to main dated 2026-09-29.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/opensumi-core)Star history chart for your README
[](https://hysenlabs.com/projects/opensumi-core)Was this useful?