RMS Memory MCP
A persistent, local, cross-IDE memory server for AI coding agents that keeps a Markdown vault outside the repository, searchable by hybrid vector and full-text search, so a project's context survives switching between editors.
RMS Memory MCP gives AI coding agents persistent, local memory that follows a project across editors such as Cursor, Zed and Claude Code, so a developer stops re-explaining architecture decisions each session or copy-pasting a stale project-instructions file between tools. Project context lives in a centralized Markdown vault outside the repository, with no configuration-file pollution inside it.
Retrieval combines embedded vector search through LanceDB with full-text search, using a multilingual embedding model for native English and Russian understanding, and an AST-aware Markdown chunker keeps code blocks and lists bound to their parent heading. Optional tree-sitter indexing covers several languages for semantic code memory, and a knowledge-graph feature tracks durable relationships between Markdown and code. A companion desktop GUI adds a visual graph editor and Git sync; the MCP server itself stays free and standalone. It is written in Rust with Tauri.