devmemory gives Claude Code and Cursor local memory on SQLite FTS5

⚡ The gist in 5 seconds - The devmemory project has released an MCP server and CLI for persistent local memory of AI agents Claude Code and Cursor. - The storage is a local SQLite database with FTS5 full-text search; the code is open in the github.com/vasikmansuri/devmemory repository. - Secret redaction and project isolation are claimed; system requirements and a version number are not specified in the description. ### 🔍 What was found In the vasikmansuri/devmemory repository, a tool has been published that the author himself describes as persistent local-first memory for AI coding agents. The offering includes two parts: an MCP server (Model Context Protocol) and a CLI. Claude Code and Cursor are explicitly named as target clients — meaning the tool connects to them as an MCP-compatible context source rather than a standalone application. The mechanics rest on four claimed capabilities. The first is SQLite FTS5: memory is stored in a local SQLite database, and the FTS5 full-text index lets the agent pull up relevant fragments on demand instead of keeping everything in context. The second is secret redaction: sensitive data is stripped from saved records. The third is project isolation: memory is partitioned by project so that one repository's context doesn't leak into another. The fourth is local-first: data is not sent to an external server — the entire database resides on the developer's machine. ### 💡 Why it matters The main pain point of AI coding agents is losing context between sessions: every new launch of Claude Code or Cursor starts from scratch, and project details have to be retold all over again. Memory MCP servers address exactly this gap, and devmemory rides the broader growth of the MCP ecosystem with two emphases: local storage (relevant for teams that can't offload their codebase externally) and per-project isolation. The delivery format — an MCP server plus a CLI rather than a plugin for a specific IDE — makes the tool portable across clients. There is no timeline for a stable release in the description: this is an open repository, not a beta program.