mxLore
A self-hosted MCP server that gives AI coding assistants a persistent, searchable knowledge base of architectural decisions, specs, plans and lessons learned, shared across projects and multiple agents.
mxLore is a self-hosted MCP server for AI-assisted software development that stores architectural decisions, specs, plans, findings and lessons learned, reached through Model Context Protocol tools from assistants such as claude.ai, Claude Code, Cursor and Windsurf. It exposes 46 MCP tools for searching, creating, updating and recalling knowledge across projects from one server managing many codebases.
Coding assistants reach all of its tools through a local proxy rather than a public endpoint or third-party connector cloud, keeping the server on localhost or the local network. It tracks which lessons-learned entries actually help and auto-tunes its recall gating, links knowledge through a graph, and lets agents exchange messages across projects through inboxes, waking a waiting assistant the moment a message arrives instead of requiring it to poll. An admin web dashboard manages team members, API keys, projects and intelligence metrics, with invite links for one-click setup, and semantic search combines vector and full-text search using MariaDB's vector support.
It requires Windows x64 and MariaDB 10.6 or newer, with Node.js needed for full hook functionality.