Community project: persistent MCP memory layer using D1 + Vectorize + Workers AI

[rahilp]

Hey

I wanted to share a community project that might be worth referencing in the examples directory or docs, since it sits squarely in the MCP + agents space this repo covers.

What it is: second-brain-cloudflare is a self-hosted MCP memory server that gives Claude, ChatGPT, Cursor, and any MCP compatible client a shared persistent memory layer across sessions. Users deploy it to their own Cloudflare account and connect it via a single Worker URL.

Stack: Workers + D1 (SQLite persistence) + Vectorize (semantic search with three-tier similarity scoring) + Workers AI (bge-small-en-v1.5 for embeddings, Llama 4 Scout for synthesis). Everything runs on the free tier.

Why it might be relevant here: It's a production MCP server implementation with some patterns not covered in the existing examples, specifically using D1 + Vectorize together for semantic memory, importance-score-weighted recall, and a nightly cron for semantic compression. Could be useful context for anyone building on top of the mcp-server or mcp-worker examples.

Traction: 277 stars, 41 forks, 3 external contributors, Obsidian plugin (161 downloads), npm CLI (172 installs/month).

Repo: https://github.com/rahilp/second-brain-cloudflare
Site: https://www.thesecondbrain.dev
Daily Stats: https://www.thesecondbrain.dev/stats

Happy to write a condensed standalone example if that would be useful to the community here, or just flagging in case it's helpful context for anyone building similar patterns.