内容
# eez-studio-mcp
An MCP (Model Context Protocol) server that lets Claude, Cursor, ZCode, DSH or any MCP client read and edit LVGL projects *inside* EEZ Studio — widget by widget, style by style — with screenshots, live checking, a wasm simulator, input injection and visual regression.
```
MCP client (Claude / Cursor / ZCode / ...)
| stdio / SSE
v
eez_mcp_server.py / mcp-server.mjs (this repo)
| HTTP (bridge on 127.0.0.1:17620)
v
EEZ Studio <-> studio-extension/ (Studio-side bridge extension)
```
> **2026-09 repo split**: the XML/UIXML toolchain, the IR compiler (`ir2eez`),
> the VS Code extension, the wasm simulator shells, examples, goldens and the
> font tooling moved to their own repo -> **[IWILLTBEST/eezml](https://github.com/IWILLTBEST/eezml)**.
> This repo now contains only the MCP line: the servers and the Studio bridge.
## What can the AI do?
| Area | Tools (selection) |
|---|---|
| Project & screens | open_project, navigate, reload_project, project overviews |
| Widget editing | read/write widget properties, styles, flags — the AI edits the project the user has open |
| Screenshots | screenshot — canvas PNG, returned to the model as an image block |
| Live checking | build -> check with errors/warnings surfaced to the model |
| Runtime debugging | debug_start (run/debug), send_input (click/swipe), read/write_variable, debug_control |
| Visual regression | visual_baseline / visual_check — golden screenshot per screen, pixel compare with AA tolerance |
### All 47 tools
| Area | Tools |
|---|---|
| IR pipeline | `read_ir`, `write_ir`, `compile`, `reload`, `navigate`, `screenshot` |
| Widget-level editing | `list_objects`, `get_object`, `update_object`, `delete_object`, `create_widget`, `create_screen`, `undo`, `redo`, `goto_object`, `get_selection` — addressing by path or objID |
| Styles & themes | `list_styles`, `update_style`, `create_style`, `delete_style`, `add_color`, `set_theme_color`, `set_preview_theme` |
| Project files | `read_project_json`, `write_project_json`, `patch_project_json` |
| Projects | `list_projects`, `select_project`, `open_project`, `create_project` |
| Assets | `list_assets`, `add_font`, `add_image` |
| Diagnostics | `read_output`, `check`, `build_project` |
| Runtime debugging | `debug_start`, `debug_stop`, `debug_control`, `debug_status`, `read_variable`, `write_variable`, `send_input` (click / swipe injection) |
| Close-ups | `screenshot_object` |
| Visual regression | `visual_baseline`, `visual_check` |
| Infra | `ping` |
Resources: project IR / schema / skill docs plus live resources `eez://checks`, `eez://debug`, `eez://state` (subscribable — pushed on change). Long operations (check / build / debug_start / add_font …) report progress.
> `visual_baseline` / `visual_check` shell out to `tools/visreg.py`, which moved to the [eezml](https://github.com/IWILLTBEST/eezml) repo in the 2026-09 split. The servers auto-detect a sibling `eezml/` checkout; alternatively point `EEZ_VISREG_SCRIPT` at it (python with PIL+numpy required, override via `EEZ_VISREG_PYTHON`).
### AI workflow manual
The step-by-step manual for AI-built projects — IR schema, layout rules, interaction patterns, font pipeline, visual-regression discipline — lives in [eezml/SKILL.md](https://github.com/IWILLTBEST/eezml/blob/main/SKILL.md) (Chinese: [SKILL.zh-CN.md](https://github.com/IWILLTBEST/eezml/blob/main/SKILL.zh-CN.md), content-equivalent).
## Screenshots
Everything below was generated from the [eezml](https://github.com/IWILLTBEST/eezml)
toolchain and captured through the MCP `screenshot` tool — the AI takes these
itself, no manual touch.
**Motor controller** (3 screens, English variant):
| overview | params | alarms |
|---|---|---|
|  |  |  |
**Glassmorphism showcase** — translucent cards, shadows, gradient bg, staggered entrance animation:
**i18n, one source two languages** — switch `strings.default` and recompile:
| English | 中文 |
|---|---|
|  |  |
**Rich data demo** — roller, gauge, calendar, spinbox, keyboard, tabview:
| main | controls | settings |
|---|---|---|
|  |  |  |
## Setup
1. **Install the Studio bridge**: import `eez-studio-mcp-extension-0.2.0.eez-extension` into EEZ Studio (it starts the HTTP bridge on port 17620).
2. **Run the MCP server**:
- Python: `pip install mcp httpx` then `python eez_mcp_server.py`
- Node: `node mcp-server.mjs`
3. Point your MCP client at it (see `claude_desktop_config.example.json`).
## Layout
| Path | What |
|---|---|
| `eez_mcp_server.py` | Python MCP server |
| `mcp-server.mjs` | Node MCP server (same bridge protocol) |
| `studio-extension/` | EEZ Studio bridge extension source |
| `tests/` | e2e MCP client test |
## License
MIT.
MCP 市场