BuildAutomata Memory MCP Server

# Build Scripts These scripts build the BuildAutomata Memory MCP extension in different formats. ## Prerequisites ### For Python Extension (build_python.sh) - Node.js 16+ (for mcpb CLI) - Python 3.9+ ### For Binary Extension (build_binary.sh) - Node.js 16+ (for mcpb CLI) - Python 3.9+ - PyInstaller (`pip install pyinstaller`) - ~2GB free disk space for build artifacts - ~10 minutes for first build ## Building ### Python Extension (Developers) ```bash ./build_scripts/build_python.sh ``` Output: `dist/buildautomata-memory-dev.mcpb` **Pros:** - Small package size (~10MB) - Users can inspect/modify source code - Fast build time **Cons:** - Requires Python 3.9+ installed on user's machine - User must install dependencies ### Binary Extension (End Users) ```bash ./build_scripts/build_binary.sh ``` Output: `dist/buildautomata-memory.mcpb` **Pros:** - Zero dependencies - works standalone - No Python installation required - Single-click install for end users **Cons:** - Large package size (50-200MB depending on platform) - Platform-specific (must build on target OS) - Slower build time (~10 minutes) ## Testing ### Local Testing (before packaging) ```bash # Test the MCP server directly cd mcpb_build/src python3 buildautomata_memory_mcp.py # Should start MCP server on stdio # Press Ctrl+C to stop ``` ### Extension Testing (after packaging) 1. Build the extension with one of the scripts above 2. Open Claude Desktop 3. Go to Settings → Extensions 4. Drag the .mcpb file from `dist/` into the extensions window 5. Click "Install" 6. Restart Claude Desktop 7. Test memory commands in a conversation ## Platform-Specific Builds ### Windows Run on Windows machine: ```bash ./build_scripts/build_binary.sh # Produces: buildautomata-memory.exe ``` ### macOS Run on Mac: ```bash ./build_scripts/build_binary.sh # Produces universal binary for Intel + Apple Silicon ``` ### Linux Run on Linux: ```bash ./build_scripts/build_binary.sh # Produces Linux ELF binary ``` ## Cross-Platform Strategy For official releases, use CI/CD (GitHub Actions) to build all platforms: - Windows x64 - macOS Universal (Intel + ARM) - Linux x64 - Linux ARM64 See `.github/workflows/build.yml` (TODO) for automated builds. ## Troubleshooting ### "mcpb: command not found" ```bash npm install -g @anthropic-ai/mcpb ``` ### "pyinstaller: command not found" ```bash pip3 install pyinstaller ``` ### Build fails with import errors Make sure all dependencies are in `requirements.txt`: ```bash pip3 install -r requirements.txt ``` ### Binary too large (>200MB) Try using Nuitka instead of PyInstaller for smaller binaries: ```bash pip3 install nuitka # See build_binary.sh for Nuitka configuration ``` ### Extension installs but doesn't work 1. Check Claude Desktop logs: `~/Library/Logs/Claude/` (Mac) or `%APPDATA%/Claude/logs/` (Windows) 2. Test MCP server directly: `python3 src/buildautomata_memory_mcp.py` 3. Verify manifest.json is valid JSON: `python3 -m json.tool manifest.json` ## File Size Estimates | Package Type | Uncompressed | Compressed (.mcpb) | |--------------|--------------|-------------------| | Python | ~10 MB | ~8 MB | | Binary | ~150 MB | ~60 MB | The binary includes: - Python runtime (~40MB) - Qdrant client libs (~30MB) - sentence-transformers (~50MB) - Application code (~10MB) - Dependencies (~20MB) ## Development Workflow 1. Make changes to source in `../` (main repo) 2. Copy updated files to `src/`: `cp ../*.py src/` 3. Test locally: `python3 src/buildautomata_memory_mcp.py` 4. Build extension: `./build_scripts/build_python.sh` 5. Test in Claude Desktop 6. If working, build binary: `./build_scripts/build_binary.sh` 7. Test binary on clean machine (no dev tools) 8. Distribute both versions ## Next Steps - [ ] Add icon.png (256x256) - [ ] Add screenshots/ for extension directory - [ ] Set up GitHub Actions for cross-platform builds - [ ] Create release workflow - [ ] Submit to Anthropic extension directory