mddb keeps your AI coding environment in sync: CLAUDE.md, skills, MCP servers, subagents. On all your machines and across harnesses. Add a rule in Claude Code on your desktop; it shows up in Codex on your laptop.
$ curl -fsSL mddb.cc/install | sh · coming soon
the problem
Claude Code and Codex each keep their own instructions. Your desktop and laptop each keep their own skills. Four environments, four different brains.
New machine or fresh install means hours of copying dotfiles, re-adding MCP servers, hunting for that one skill you wrote last month.
An agent rewrites its own config, something breaks, and there is no history, no diff, no rollback. You reconstruct it from memory.
Rules in CLAUDE.md, skills in dotfiles, MCP servers in JSON, env vars in your head. One environment, scattered across files, folders and formats.
An MCP server sits in the config but stopped working weeks ago. Nothing checks it, nobody notices until it fails mid-demo.
Five developers, five different setups. The practices everyone agreed on never quite make it into everyone's agents.
what mddb does
mddb doesn't copy files between machines. It parses your config into meaning (rules, skills, MCP servers, subagents), versions every change, and compiles it back into each harness's native format.
mddb switch work swaps rules, skills, MCP servers and subagents across every harness on the machine. Personal, work, a profile per client: each one is a complete environment, versioned and reversible. Like branches, but for your AI setup.
One rule becomes a managed section in CLAUDE.md for Claude Code and in AGENTS.md for Codex. Edit either side and it syncs back.
No background services. The sync engine lives inside an MCP server your harness starts with each session. Your agent can manage its own config by talking to it.
mddb login && mddb restore. A fresh install becomes your fully configured environment in under two minutes.
Every change is an immutable event. See who changed what, from which machine, and roll any of it back.
Different release skills on work and home machines? Pin any skill to specific devices. Sync doesn't mean making every machine identical.
API keys and tokens are never uploaded, only references and env-var names. Built-in secret scanning blocks accidents before they sync.
how it works
mddb init scans your machine, finds Claude Code and Codex, and imports rules, skills and MCP configs into your profile. You pick what syncs.
Edit configs by hand, by CLI, or let your agent do it via MCP tools. Changes propagate to every connected machine and harness in seconds.
New laptop, broken config, wiped disk: mddb restore brings back your exact environment, with a safety backup of whatever was there.
mddb core · open source
Everything that touches your machines is open source: the sync engine, the MCP host, the CLI, every adapter and the config schema. Self-host the server with no limits, audit every line, export everything and leave anytime. Core is the foundation both Orbit and Fleet stand on.
mddb orbit · pricing, early & negotiable
Orbit is mddb Core hosted for you: zero setup, backups, history, support. Prefer your own server? Core is free and unlimited, forever. Pick your mass and upgrade as your gravity grows.
// beta pricing: numbers are a starting point for a conversation, not a wall. need a whole universe? meet mddb fleet
principles
We're building mddb in the open. Get on the list and be first to sync.
Join the waitlist