opencode-memory 0.4.3

Local Memory MCP Server for OpenCode

Memory MCP

Local Memory MCP Server for OpenCode - A persistent memory layer for AI coding agents.

Features

• Local-only memory storage (no external services except embeddings)
• FAISS vector search for semantic memory retrieval
• SQLite for metadata persistence
• MCP protocol for integration with OpenCode
• Concurrent access via file locking

Quick Start

# Set environment variables
export MEMORY_USER_ID=your_user_id
export EMBEDDING_PROVIDER=openai  # or ollama, huggingface
export EMBEDDING_API_KEY=your_api_key

# Run the server
python -m memory_mcp.server

OpenCode Configuration

Add to your OpenCode MCP configuration file:

{
  "mcp": {
    "memory": {
      "type": "local",
      "enabled": true,
      "command": ["uv", "run", "python", "-m", "memory_mcp.server"],
      "environment": {
        "MEMORY_USER_ID": "your_user_id",
        "MEMORY_DATA_DIR": "/path/to/data",
        "EMBEDDING_PROVIDER": "openai",
        "EMBEDDING_API_KEY": "your_api_key"
      }
    }
  }
}

Or after installing via pip:

uvx opencode-memory

#!/bin/bash
export MEMORY_USER_ID=your_user_id
export MEMORY_DATA_DIR=./data
export EMBEDDING_PROVIDER=openai
export EMBEDDING_API_KEY=your_api_key
exec python -m memory_mcp.server

Environment Variables

Variable	Required	Description
MEMORY_USER_ID	Yes	User/entity ID for operations
MEMORY_DATA_DIR	No	Base directory for data (default: ./data)
EMBEDDING_PROVIDER	Yes	openai, ollama, or huggingface
EMBEDDING_API_KEY	No	API key for embedding service
EMBEDDING_BASE_URL	No	Custom endpoint for Ollama/HF
EMBEDDING_MODEL	No	Model name
EMBEDDING_DIMENSION	No	Vector dimension (default: 1536)