MLView plugin
A plugin for Claude Code, Codex and Copilot that maps ML training code into a diagram linked to source lines
- 428 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill reads source, notebooks and configuration and writes a diagram file into the workspace
- A Python helper runs on your machine
Install
- Works in
Manual install
/plugin marketplace add realmyang/MLViewThen run /plugin install mlview-github@mlview-local. The plugin is taken from the claude-plugin folder on main.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill reads source, notebooks and configuration and writes a diagram file into the workspace
- A Python helper runs on your machine
This is third-party code. Review the repository files before installing.
What it does
MLView is a plugin with one mlview skill that asks the agent's active model to analyse an ML project and publish a diagram as a `*.mlview.json` file. A VS Code extension comes with it and opens that file as an interactive diagram. No separate API key is needed: the model, tools and permissions come from your assistant. The agent reads source, configuration, notebooks, launch scripts and tests, picks a scenario and describes data, model, losses, parameter updates and validation. Every step and connection carries quotes from the code, and every claim is marked observed, inferred or unresolved. A helper script validates the file structure and that quotes match the lines, but it never imports or runs the analysed project. In the editor, selecting a step opens the quoted lines beside the diagram. Right-clicking a code line jumps to the step that cites it. To refine, you copy a prepared prompt, and a new revision updates the panel.
Who it is for. For ML engineers and researchers who study someone else's or an old training project and want the path of data and losses traced to code.
Good fit when
- You inherited a training repository and need to see how data reaches the loss function
- Before editing a config you want a diagram of stages: data, model, optimizer, validation
- A review needs suspicious spots traced with quotes and an "inferred" or "unresolved" mark
- You need to confirm quotes in the diagram still match the code after files changed
- Notebooks and configs need one picture: the agent accounts for notebook cells and launch arguments
Not a fit when
- You need proof that the model's interpretation is right: quotes are matched to lines, the meaning is for you to judge
- You need to run or profile training: the plugin only reads code
- The project is not about machine learning: the skill is tuned for data flow, losses and parameter updates
- You need a ready extension from a marketplace: today the VSIX is built from source
Example request
Explain training with this config. Show data, model, losses, parameter updates, validation, and anything unresolvedLimitations
The implementation is experimental, and the author tracks validation status in docs/STATUS.md. The VS Code extension is installed from a source build: it needs Node.js 20.18.1 or newer and Python 3.10 or newer, then a VSIX install through the editor command. The Claude Code plugin is installed separately. The diagram content and findings are written by the active model, so quote validation does not prove the interpretation is correct. The marketplace is named mlview-local whatever the source, so add it from one place only.
How to disable. In Claude Code run `/plugin uninstall mlview-github@mlview-local`. For Codex and Copilot projects the skill sits in `.agents/skills/mlview/`; delete that folder. Remove the extension in the VS Code extensions list.
FAQ
How do I install MLView in Claude Code?
Run `/plugin marketplace add realmyang/MLView`, then `/plugin install mlview-github@mlview-local`.
Do I need a separate API key?
No. The analysis is done by the model already connected to your assistant: Copilot, Codex or Claude Code. The local helper only validates file structure and quotes.
Does MLView run my project?
No. The skill reads files with the agent's tools, and the Python helper never imports or executes the analysed program. A line in a test or notebook does not mean it ran.
How do I check a specific diagram step?
Press `r` or the Review button: the walk starts with steps marked inferred or unresolved. Each step selects a claim and, after a short pause, opens the quoted lines beside the diagram.
What if a file cited in the diagram has changed?
The status bar shows whether cited files are unchanged. If a file changed, the walk offers the claims that cite it and does not open the stale quote.
Discussion
No questions or reviews yet.
Related
PluginsA Claude Code plugin and MCP server: over 1000 scientific tools for biology, chemistry and medicine, data-check commands and routing skills
A skill set for math modeling contests: problem parsing, method selection, Python and MATLAB code, and paper writing
A skill that turns SMILES into a Mermaid graph of named atoms, so the agent edits molecules without hand-writing SMILES
MCP server that connects Claude Code and Codex to an open R or Python session in RStudio and Positron, where your data lives
A CLI for fine-tuning language models from one YAML: LoRA, QLoRA, DPO and GRPO, layer streaming for 8B on a 4 GB GPU, export to GGUF
MCP server and CLI to drive real Microsoft Excel: Power Query, DAX, VBA, PivotTables and charts
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| MLView (this entry) | no | 428 | |
| visual-explainer | no | 10,692 | builds HTML pages with diagrams, diff reviews and slides for any code |
| Diagram Design | no | 62,871 | draws diagrams of any system as HTML and SVG, with no code quotes or confidence marks |
README in short
The project has three parts: a skill for the assistant, a shared diagram file format and a VS Code viewer. Most of the documentation covers the viewer, with its claim walk, revision comparison and phase overview.