MemMCP

MemMCP is a hyper-optimized Memory Server natively implementing the Model Context Protocol (MCP). It bridges the gap between lexical keyword search and semantic vector embeddings, delivering 100% deterministic, deduplicated memory recall for autonomous AI Agent Swarms.

📖 Table of Contents

• Why MemMCP?

• Core Features

• Quick Start

• Architecture

• MCP Tool Reference

• Contributing & Security

• License

Related MCP server: agent-memory

🤔 Why MemMCP?

When dozens of autonomous agents operate in parallel, standard vector databases suffer from race conditions, data duplication, and context hallucination. MemMCP solves this by merging SQLite Write-Ahead Logging (WAL) for ACID-compliant state management with FAISS Hybrid Reciprocal Rank Fusion (RRF) for unparalleled semantic retrieval.

✨ Core Features

🚀 Quick Start

MemMCP is designed to be booted instantly by any MCP-compliant LLM or Agent Framework via standard IO streams (stdio).

# 1. Clone the repository
git clone https://github.com/axton/project_2_mcp_memory.git
cd project_2_mcp_memory

# 2. Build the exact dependency graph using uv
make build

# 3. Verify the rigorous mathematical test suite
make test

# 4. Boot the MCP server directly
make run

🏗 Architecture

graph TD
    A[MCP Client] --> B{Bloom-Filter Idempotency Gate}
    B -->|Duplicate Request| C[Drop (Idempotent Return)]
    B -->|New Request| D[Vectorization]
    D --> E[FAISS Hybrid RRF Search]
    E --> F[SQLite WAL Merkle-Root Ledger]
    F --> G[XML RAG Formatter]
    G --> H[Response]

🛠 MCP Tool Reference

MemMCP automatically exposes the following functions to any connected agent:

• store_memory: Store a single memory. Generates unique keys and updates FAISS indices.

• store_memories_batch: Store multiple memories atomically in a single massive transaction, rebuilding the index only once.

• recall_memories: Retrieve relevant memories using Reciprocal Rank Fusion (blending Semantic FAISS similarity + SQLite FTS5 keyword search).

🤝 Contributing & Security

To contribute, you must abide by our strict mathematical isolation limits. See CONTRIBUTING.md for details. For vulnerabilities, refer to SECURITY.md.

📄 License

MIT License. Copyright (c) 2026 Axton Carroll.