Memory MCP Server
Persistent knowledge graph for LLM workflows
A cloud-based persistent memory system that gives AI coding assistants long-term memory across sessions, devices, and projects. Built on AWS serverless infrastructure with semantic search powered by Bedrock embeddings.
Features
Knowledge graph storage
Single-table DynamoDB design with adjacency list pattern. Stores entities (projects, decisions, bugs, concepts, people) and their relationships in one table.
Semantic search
Bedrock Titan Embed V2 generates 256-dim vectors stored alongside entities. Natural-language queries find relevant knowledge via cosine similarity.
Session tracking
Chronological session records — what was done, what was decided, which entities were involved. Filter by project, sorted newest-first.
MCP tools
Six tools via MCP: memory_write, memory_search, memory_graph_query, memory_list_sessions, memory_get_session, memory_get_context. Plug into any MCP client.
S3 document archive
Every entity backed up as human-readable Markdown in S3. Browse your knowledge base as files, or use it as a future Bedrock Knowledge Base source.
Architecture
API Gateway → Lambda (FastMCP + Mangum) → DynamoDB (entities + sessions) + S3 (markdown archive) + Bedrock (embeddings). Fully serverless, pay-per-request.