opencode-personal-knowledge

This MCP server provides a personal knowledge base with local embeddings and a vector store. It lets you store knowledge entries, perform semantic searches, and manage entries with a suite of knowledge tooling. It also supports session-based memory for OpenCode sessions, enabling you to log messages and search across sessions. The tooling includes semantic search, text search, and management commands to store, update, retrieve, list, and delete knowledge entries, as well as session logging and retrieval features. All processing runs locally, using an embedded vector database and local embeddings, so there are no external API keys required.

Prerequisites:

• Bun installed (bun.sh)
• Basic Node.js/Bun environment

Install and run the MCP server:

# 1. Clone the repository
git clone https://github.com/NocturnLabs/opencode-personal-knowledge.git
cd opencode-personal-knowledge

# 2. Install dependencies
bun install

# 3. Start the MCP server
bun run mcp

Usage (during development):

• The MCP server starts with the command above and exposes tools you can invoke via the MCP framework. By default, data is stored under ~/.local/share/opencode-personal-knowledge/ and embeddings are generated locally via FastEmbed with LanceDB for vector storage.

Environment configuration (optional):

• You can customize the data directory by setting OPENCODE_PK_DATA_DIR, e.g.:

export OPENCODE_PK_DATA_DIR=/custom/path

Notes and tips:

• Data location: By default, knowledge and vector data are stored under ~/.local/share/opencode-personal-knowledge/; override with OPENCODE_PK_DATA_DIR.
• Embeddings: Uses BGE-small-en-v1.5 with FastEmbed (auto-downloads on first use).
• Runtime: The MCP server is run via Bun (bunx). Ensure Bun is installed and available in your PATH.
• Embeddings and vector store are fully local; no external API keys are required.
• If you encounter permission issues writing to the data directory, check directory permissions and adjust OPENCODE_PK_DATA_DIR accordingly.