goldie

Goldie is a retrieval-augmented generation (RAG) MCP server written in Go that runs locally on your machine. It supports multiple embedding backends, including MiniLM via ONNX Runtime and Ollama, with a SQLite-backed vector store for persistent indexing and semantic search. You can index content, files, and directories, then perform semantic queries to retrieve relevant documents or recall knowledge with source attribution. The server exposes tools for indexing content (index_content, index_file, index_directory) and for querying and managing the index (search_index, recall, delete_document, count_documents). To run Goldie, start the goldie-mcp binary and point Claude or your MCP client to the server, using the environment variables described in the README to select the embedding backend and storage location. The default behavior uses MiniLM on ONNX Runtime unless you specify Ollama via the CLI flag -b ollama or equivalent configuration.

Prerequisites:

• Go 1.22+ installed on your system
• Git installed
• If building from source, Git LFS is required because the model file is stored with LFS
• Optional: ONNX Runtime library if using MiniLM backend and you plan to run locally

From Releases (recommended):

1. Download the appropriate pre-built binary from the releases page:
   • macOS Apple Silicon: goldie-mcp-darwin-arm64
   • macOS Intel: goldie-mcp-darwin-amd64
   • Linux x86_64: goldie-mcp-linux-amd64
   • Linux ARM64: goldie-mcp-linux-arm64
2. Make the binary executable and relocate it to your bin directory: curl -LO https://github.com/srfrog/goldie-mcp/releases/latest/download/goldie-mcp-darwin-arm64 chmod +x goldie-mcp-darwin-arm64 mv goldie-mcp-darwin-arm64 ~/bin/goldie-mcp

From Source (requires Go and Git LFS):

1. Install Git LFS (if not already configured): git lfs install
2. Clone the repository and build: git clone https://github.com/srfrog/goldie-mcp cd goldie-mcp make build
3. The build outputs the goldie-mcp binary which you can place in your PATH.

Configuration notes:

• If you plan to use Ollama, ensure Ollama is installed and running, and you may need to pass -b ollama to the binary or configure the equivalent in your environment. For Ollama integration, set OLLAMA_HOST, OLLAMA_EMBED_MODEL, and possibly OLLAMA_EMBED_DIMENSIONS.
• If you plan to use MiniLM locally, ensure ONNX Runtime libraries are installed and accessible, and set ONNXRUNTIME_LIB_PATH if needed.
• The default storage location is GOLDIE_DB_PATH pointing to a sqlite database. Change it as needed.

Tips and common issues:

• When indexing directories, Goldie will automatically skip common patterns (dotfiles, node_modules, vendor, etc.) unless overridden by a .goldieskip file.
• If using Ollama, ensure the selected Ollama embedding model is pulled (e.g., nomic-embed-text) and that Ollama is running before starting Goldie with -b ollama.
• For Claude integrations, you can embed Goldie in your workflow via the provided mcp configuration examples, including env settings for GOLDIE_DB_PATH and embedding backend selections.
• If ONNXRUNTIME_LIB_PATH is not set, the runtime library will be auto-detected; however, you may need to provide the path on some systems.
• When using search_index or recall, you can adjust limits and depth via the tool parameters to tailor result size and source attribution.