memory-mcp
Verified Safeby michael-denyer
Overview
Provides Claude with persistent 'two-tier' memory (hot cache for zero-latency access, cold storage for semantic search) and automatic pattern mining from outputs to enhance long-term context and reduce token usage.
Installation
uv run --directory /path/to/memory-mcp memory-mcpEnvironment Variables
- MEMORY_MCP_DB_PATH
- MEMORY_MCP_EMBEDDING_MODEL
- MEMORY_MCP_EMBEDDING_DIM
- MEMORY_MCP_EMBEDDING_BACKEND
- MEMORY_MCP_HOT_CACHE_MAX_ITEMS
- MEMORY_MCP_PROMOTION_THRESHOLD
- MEMORY_MCP_DEMOTION_DAYS
- MEMORY_MCP_AUTO_PROMOTE
- MEMORY_MCP_AUTO_DEMOTE
- MEMORY_MCP_LOG_RETENTION_DAYS
- MEMORY_MCP_DEFAULT_RECALL_LIMIT
- MEMORY_MCP_DEFAULT_CONFIDENCE_THRESHOLD
- MEMORY_MCP_HIGH_CONFIDENCE_THRESHOLD
- MEMORY_MCP_TRUST_SCORE_MANUAL
- MEMORY_MCP_TRUST_SCORE_MINED
- MEMORY_MCP_MAX_CONTENT_LENGTH
- MEMORY_MCP_MAX_RECALL_LIMIT
- MEMORY_MCP_MAX_TAGS
Security Notes
The server explicitly states it is for 'local use only' and runs unauthenticated over STDIO, warning against exposure to networks. Input content is validated for length (`max_content_length`) to prevent excessive resource consumption. SQLite queries use parameterized statements to prevent SQL injection. External ML models are downloaded from trusted sources (Hugging Face, mlx-community), which introduces a dependency risk, but is common practice. No obvious malicious patterns or hardcoded secrets were found.
Similar Servers
claude-memory-mcp
Provides local, persistent, and searchable memory for AI assistants like Claude Desktop, enabling them to store and recall information efficiently.
claude-conversation-memory-mcp
Provides long-term memory for AI coding agents by indexing conversation history, tracking decisions and mistakes, and enabling semantic search across projects.
post-cortex
Provides long-term, persistent memory and knowledge management for AI assistants, enabling them to store, semantically search, and retrieve conversation context, decisions, and code-related insights.
mcp-structured-memory
Provides structured, domain-specific memory management for AI agents to use in ongoing projects, storing accumulated context in local markdown files.