Gemini Embedding 2 Mcp
@AlaeddineMessadi
A powerful Model Context Protocol (MCP) server using gemini embedding 3 that transforms any local directory into an ultrafast, visually-aware spatial search engine for AI agents.
Overview
What is Gemini Embedding 2 Mcp?
Gemini Embedding 2 Mcp connects local documents, code, images, and videos to Claude, Cursor, or VS Code using Google's gemini-embedding-2-preview model and a strictly local ChromaDB vector database. It is designed for users who need local-first, enterprise-grade semantic search and RAG with full privacy.
How to use Gemini Embedding 2 Mcp?
Install uv (pip install uv), get a free Gemini API key from Google AI Studio, then configure your client using the uvx command with the GEMINI_API_KEY environment variable. Zero-Install (recommended) runs directly from GitHub without cloning; alternatively, clone the repository and run uv sync for local development.
Key features of Gemini Embedding 2 Mcp
- Local privacy with ChromaDB stored at
~/.gemini_mcp_db - Enterprise‑grade embeddings via
gemini-embedding-2-previewwith MRL 768 optimization - Native support for images, video, and audio without text extraction
- Visual PDF RAG page‑by‑page preserving charts and layout
- Agentic guardrails: junk filter, wildcard blacklisting, exponential backoff, ghost pruning
Use cases of Gemini Embedding 2 Mcp
- Index a local codebase and perform semantic searches across source files
- Run visual RAG on scanned PDFs containing charts and embedded plots
- Search over images, videos, and audio files without manual transcription
- Let an autonomous AI agent safely index and sync directories with automatic blacklisting
FAQ from Gemini Embedding 2 Mcp
Where does my data and embeddings live?
All vectors and metadata are stored locally in ChromaDB under ~/.gemini_mcp_db. Only raw byte chunks are sent to the Gemini Embedding API; your files never reach a third‑party database.
What model and dimensionality are used?
It uses the gemini-embedding-2-preview model with the RETRIEVAL_DOCUMENT task type and MRL dimensionality of 768.
What file types are supported?
Images (.jpg, .webp), video (.mp4), audio (.mp3, .wav), PDFs (visual RAG), and any text/code files present in indexed directories.
How do I install it without cloning the repo?
Use the zero‑install method via uvx. Your client (Claude Code, Claude Desktop, Cursor, etc.) points to uvx --from git+https://github.com/AlaeddineMessadi/gemini-embedding-2-mcp-server.git gemini-embedding-2-mcp with the GEMINI_API_KEY environment variable.
What MCP tools are exposed?
index_directory, search_my_documents, list_indexed_directories, sync_indexed_directories, remove_directory_from_index, and the resource gemini://database-stats for real‑time ChromaDB observability.
