Early access — invite only

About EngramMCP

EngramMCP is a local-first semantic memory layer for AI agents. It exposes a Model Context Protocol (MCP) server that lets any AI agent — Claude Code, Cursor, custom runtimes — capture, search, and enrich memory across multiple typed sources (notes, conversations, Drive, Notion, YouTube, Obsidian, audio).

How it works

The MCP server runs locally on your PC. Your memories (text + 768-dim vector embeddings) live in a local SQLite + LanceDB database — they never leave your machine in clear.

If you upgrade to Pro, your local engram-mcp can sync encrypted ops to our cloud (Cloudflare, EU-hosted). That gives you mobile access, multi-PC sync, and an always-online dashboard at engram-mcp.com — without us ever seeing your content. Encryption is E2E (XChaCha20-Poly1305) with a passphrase-derived key (Argon2id) that lives only on your devices.

Philosophy

Local-first: Your memory lives on your PC. The cloud is optional, only for sync/transit when you opt in to Pro.
Privacy by design: Default embeddings + transcription happen locally (Ollama, whisper.cpp). Pro users can opt into hosted compute if they prefer (with quota + transparent overage pricing).
No hidden lock-in: Your memory store is a regular SQLite file. Export at any time.
Open core: The MCP server is OSS. The cloud workers (sync, hosted AI) are proprietary.

Built by

EngramMCP is built by RavioleLabs — a software studio building tools for developers and AI power users.

Get an invite

Engram is in early access. Join the RavioleLabs Discord to request access, chat with the team, and follow updates.