Engram MCP — DevOps skill for Claude Code

🧠 Engram MCP

**A zero-dependency, auto-decaying, pure SQLite MCP server for persistent AI agent memory.**

*Built in an afternoon. Tested harder than it was built. Documented because the code isn't enough.*

*engram (noun): a hypothesized physical trace of memory stored in the brain — the biological basis of how memories persist.*

[](LICENSE) [](https://python.org) [](#) [](https://modelcontextprotocol.io) [](https://github.com/lalithbuilds/engram-mcp/actions/workflows/ci.yml)

Engram is a fully local, lightning-fast memory layer for AI agents (Claude Code, Cursor, Windsurf, etc.) connected via the [Model Context Protocol (MCP)](https://modelcontextprotocol.io). It is built entirely on the Python 3 standard library. No cloud API keys, no vector databases, no Docker, and no bloat.

---

🛑 The Problem: "Agent Amnesia"

If you use AI coding agents, you know the frustration: **every session starts completely fresh**. The agent doesn't remember that you prefer `pnpm` over `npm`, it forgets the architectural boundaries you agreed on yesterday, and it constantly asks you for the same context.

Existing solutions try to fix this by bolting on massive Vector Databases (PostgreSQL/pgvector, Chroma, Pinecone) that require heavy dependencies, Docker containers, and expensive OpenAI API calls just to generate embeddings.

💡 The Solution: What is Engram MCP?

**Engram MCP** is a ruthlessly optimized, zero-dependency alternative. It uses standard **SQLite FTS5 (Full-Text Search)** to achieve blazing-fast keyword retrieval entirely locally. You simply drop the `server.py` script into your MCP configuration, and your