Files
funplay-cocos-mcp/CONTRIBUTING.md
T
mingyuansi b65a4f22c3 fix: Windows spawn, prefab path, preview API, asset deps, error messages; add execute-context resource, tool descriptions, scene/script/asset tools
Bug fixes (verified against Cocos Creator 3.8.8):
- diagnostics: shell:true for .cmd/.bat on Windows (T125/T126)
- prefabs: duplicatePrefab target resolves under assets/ (T420)
- cocos-project: add preview.open candidate for 3.8.8 (T444)
- assets-advanced: detect directory assets in inspectAssetDependencies (T110)
- scene: improve component-not-found errors with compilation hint (T429)

Documentation improvements:
- tool-registry: add enum and injected vars to execute_javascript, path format examples, asset ref limitation note
- resources: add cocos://mcp/execute-context resource with variables, patterns, pitfalls

New tools (core 37->38, full 101->110):
- scene-management: create_scene, query_scene_state (core), copy_paste_node, rename_node, reparent_node
- scripts: create_script with component/plain templates
- prefabs: create_prefab
- assets-advanced: batch_asset_ops, find_unused_assets
2026-06-30 21:53:06 +08:00

142 lines
4.9 KiB
Markdown

# Contributing to Funplay MCP for Cocos
Thanks for your interest in contributing to Funplay MCP for Cocos.
This repository is a Cocos Creator editor extension that embeds an MCP server. Contributions that improve reliability, editor compatibility, documentation, and AI-driven Cocos workflows are welcome.
## Development Setup
1. Clone the repository:
```bash
git clone https://github.com/FunplayAI/funplay-cocos-mcp.git
cd funplay-cocos-mcp
```
2. Install it into a Cocos project as an extension:
```bash
mkdir -p /path/to/your-cocos-project/extensions
ln -s "$PWD" /path/to/your-cocos-project/extensions/funplay-cocos-mcp
```
You can also copy the repository folder instead of using a symlink.
3. Restart Cocos Creator or reload extensions.
4. Open the panel:
```text
Funplay > MCP Server
```
## Validation
Before opening a pull request, run:
```bash
npm run check
```
This validates the JavaScript syntax for the extension entry points and helper modules.
If your change affects live MCP behavior, also test against a running Cocos project:
```bash
curl -sS http://127.0.0.1:8765/health
```
If you use a custom port, replace `8765` with your configured port.
## Project Structure
```text
browser.js Cocos browser/editor extension entry and service lifecycle
scene.js Scene-side execution bridge and scene tool implementations
panel/index.js Minimal MCP Server panel UI
lib/server.js HTTP JSON-RPC MCP server
lib/tool-registry.js Tool definitions and handlers
lib/resources.js MCP resources
lib/prompts.js MCP prompts
lib/client-config.js One-click MCP client configuration
lib/assets.js Asset database helpers
lib/diagnostics.js TypeScript diagnostics
lib/input.js Electron input simulation
lib/screenshots.js Screenshot helpers
```
## Contribution Guidelines
- Keep the `core` profile focused. Avoid adding noisy tools to `core` unless they are broadly useful for AI clients.
- Prefer improving `execute_javascript` workflows when one flexible tool is better than many narrow tools.
- Add tools to `full` when they are useful but not essential for the default AI workflow.
- Keep the Cocos panel simple. The main panel should focus on service management and MCP client configuration.
- Avoid adding runtime dependencies to built games; this extension should remain editor-only.
- Keep changes small and focused.
- Update `README.md`, `README_CN.md`, or `CHANGELOG.md` when behavior changes.
- Do not commit generated Cocos folders such as `library/`, `temp/`, or build output.
## Tool Design Notes
When adding a new tool:
1. Add the tool definition in `lib/tool-registry.js`.
2. Use `profile: "core"` only for high-signal tools.
3. Put scene/runtime work behind `sceneBridge.call(...)` when it must run in the active Cocos scene.
4. Use editor-side helpers for asset-db, filesystem, and panel/client configuration workflows.
5. Return JSON-serializable data where possible.
6. Add clear input schema descriptions so AI clients can choose the tool correctly.
7. Test the tool with `tools/list` and `tools/call`.
### Asset Creation via Template Files
When a tool creates a new Cocos asset with a complex serialized format
(scene, prefab, animation clip, material, etc.), **store a template file in
`resources/` and pass its content to `asset-db: create-asset`** rather than
hard-coding JSON strings in source.
Why:
- Cocos serialized formats are version-specific and contain many required
sub-objects (e.g. a scene needs `cc.SceneGlobals``cc.AmbientInfo`,
`cc.ShadowsInfo`, `cc.SkyboxInfo`, `cc.FogInfo`, `cc.OctreeInfo`,
`cc.SkinInfo`, `cc.LightProbeInfo`, `cc.PostSettingsInfo`). Missing or
misnamed classes cause silent import failures or runtime errors.
- A real template file can be created/verified in the Cocos editor and
updated by simply replacing the file.
- `asset-db: create-asset(url, content)` handles `.meta` generation and
import in one step — no need for `fs.writeFileSync` + `refreshAssets`.
Pattern:
1. Create the template file at `resources/template.<ext>` (e.g.
`resources/template.scene`).
2. Add `"resources/"` to the `files` array in `package.json` so it ships
in the npm package.
3. In the tool handler, read the template with `fs.readFileSync`, optionally
replace `_name` fields, then call `create-asset` with the content string.
4. Export a pure helper (e.g. `buildSceneContent`) for unit testing.
## Documentation
The documentation is bilingual:
- `README.md` for English
- `README_CN.md` for Chinese
When adding user-facing behavior, update both files when possible.
## Pull Requests
A good pull request should include:
- A clear summary of the change
- What Cocos Creator version was tested
- Whether the MCP server was tested through an AI client or direct HTTP calls
- Any screenshots if the change affects the panel or visual scene output
- Notes about compatibility or limitations
## License
By contributing, you agree that your contributions will be licensed under the MIT License used by this repository.