Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@Local Search find where API keys are validated in my indexed code"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A native macOS app and local MCP service for searching user-authorized code, documents, and images with EmbeddingGemma 2. One background service shares an offline model across folders and assistants.
Requires a Mac with Apple Silicon and macOS 14 or newer.
Download Local-Search-<version>.dmg from the latest release and drag Local Search to Applications.
Open it. The app is not notarized by Apple yet, so macOS blocks the first launch: open System Settings → Privacy & Security and choose Open Anyway.
Click Install. The app downloads its search engine (Python 3.12 and libraries, about 1.4 GB) and the EmbeddingGemma 2 model (1.5 GB) once. If you already have the model, choose Use a folder I already have… instead of downloading it again.
After setup, indexing and search run offline. Everything the app installs or stores lives in ~/Library/Application Support/Local Search/; to uninstall, delete the app and that folder.
Requires uv and Xcode command-line tools.
uv sync --extra dev
.venv/bin/python scripts/build-mac-app.py # standalone app, same first-launch setup
open '.code-search/Local Search.app'For development, .venv/bin/code-search download-model puts the model in models/embeddinggemma-2/ and scripts/build-mac-app.py --dev links the app to this checkout instead; see development.
Choose folders and file types in the app, then search with your own words. No folder is authorized by default. Settings / Réglages (Command-comma) controls the interface language, precision, dimensions, text limits and image encoding. Connect an assistant / Connecter un assistant provides MCP commands and project-specific grants. The interface follows your macOS language preference by default, with English as the fallback; you can also select a language in Settings.
Indexing and search computation stay local; snippets and image previews requested by a cloud assistant can be sent to its provider. Network access is used only during setup, as described in privacy.
Related MCP server: Apple Tools MCP
Start with the documentation index. Detailed behavior, commands and limitations live under docs/.
Topic | Guide |
Choosing folders, search and previews | |
Precision, dimensions and other model options | |
Connecting an assistant and project access | |
Workspace and single-repository services | |
Code map, indexing and retrieval | |
Boundaries, revocation and stored data | |
Setup, builds and contribution checks | |
Tests, smoke checks and measured results |
Coding agents should read AGENTS.md. Guidance for assistants working in indexed projects is in docs/agent-instructions.md.
Apache-2.0. The EmbeddingGemma 2 checkpoint is downloaded separately and is also distributed under Apache-2.0.
This server cannot be deployed
Apple Developer Documentation with Semantic Search, RAG, and AI reranking for MCP clients
Persistent memory and knowledge graph for AI assistants — keyword + vector + graph search.
Search your knowledge bases from any AI assistant using hybrid RAG.
Project memory, semantic code search, and grounded agent context.
