Overview
What is Vector Memory MCP Server?
Vector Memory MCP Server is a secure, vector-based memory server for Claude Desktop that uses sqlite-vec and sentence-transformers. It provides persistent semantic memory capabilities to enhance AI coding assistants by remembering and retrieving relevant coding experiences, solutions, and knowledge.
How to use Vector Memory MCP Server?
Install via uvx (recommended) or from source. Configure Claude Desktop by adding a JSON entry pointing to the server and a working directory. Then use available tools such as store_memory, search_memories, list_recent_memories, get_memory_stats, clear_old_memories, get_by_memory_id, and delete_by_memory_id via natural language commands.
Key features of Vector Memory MCP Server
- Semantic search using 384-dimensional embeddings
- Persistent SQLite storage with vector indexing
- Smart organization with categories and tags
- Input validation, path sanitization, and resource limits
- Fast embedding generation with
sentence-transformers - Automatic deduplication via SHA-256 content hashing
- Access tracking with counts and timestamps
- Smart cleanup algorithm based on recency and importance
Use cases of Vector Memory MCP Server
- Store and retrieve code patterns, bug fixes, and architecture decisions
- Maintain team conventions, deployment procedures, and infrastructure knowledge
- Capture learning insights, performance discoveries, and security learnings
- Enable semantic search across past coding experiences and solutions
FAQ from Vector Memory MCP Server
What is the technical stack of Vector Memory MCP Server?
It uses sqlite-vec for vector storage and similarity search, the sentence-transformers/all-MiniLM-L6-v2 model for 384-dimensional embeddings, and the FastMCP framework.
What are the security limits?
Maximum memory content length is 10,000 characters, maximum total memories is 10,000 entries, maximum search results per query is 50, and maximum tags per memory is 10. Path validation blocks suspicious characters.
How does semantic search work?
Memories are converted into 384-dimensional vectors that capture semantic meaning. Search queries are embedded and compared using similarity scoring (0.6–1.0 range) to find relevant matches.
How can I install Vector Memory MCP Server?
You can install via uvx (recommended), from source using uv, or with pipx. Publishing to PyPI is in progress.
What dependencies are required?
Python 3.10 or higher (recommended 3.11), the uv package manager, and the Claude Desktop app.
