Open source · Claude Code & Cowork
More than thirty Claude Code plugins, built for our own agent fleet and published for everyone: MCP servers, skills and scheduled routines that run real operations work — from WhatsApp and Google Drive to time tracking, session backups and keeping the Macs that run the agents healthy. Free, MIT-licensed, maintained in public.
Browse the marketplace on GitHub Install in two commandsQuick facts. Marketplace name msapps-plugins, repository MSApps-Mobile/claude-plugins, 30+ plugins, declaring their SOSA compliance level (Supervised, Orchestrated, Secured Agents) in their manifests. Works in Claude Code and in Cowork. Built and maintained by a team led by a Claude Certified Architect.
In Claude Code, add the marketplace once, then install any plugin by name:
/plugin marketplace add MSApps-Mobile/claude-plugins
/plugin install mac-agent-hygiene@msapps-pluginsIn Cowork: Settings → Plugins → Marketplaces → Add → MSApps-Mobile/claude-plugins, then search for the plugin and press Install.
A Mac that runs Claude Code agents all day fills its disk in places no cleanup app looks. On one of our own machines, free space fell to 16 GB of 926 — and Downloads, Documents and Desktop together held 2 GB. The space was in 774 git worktrees agents had created per task and never removed, 937 leftover plugin-install clones, browser-profile caches and Xcode build output. A daily routine recovered about 200 GB in a morning. mac-agent-hygiene is that routine as a plugin, so any Mac can install it.
The MCP server exposes hygiene_status, hygiene_plan, hygiene_apply and hygiene_schedule. A plan lists every candidate with remove or keep and the reason; apply re-checks each item and refuses anything that changed. Nothing is deleted without a plan id and an explicit confirm.
Xcode DerivedData (never during a build), leftover plugin clones older than a day, the regenerable cache folders inside Playwright browser profiles, re-downloadable model caches, and stale git worktrees.
Browser logins — cookies, local storage, IndexedDB stay. Any worktree that is dirty, unpushed, in use or younger than a week. Anything outside your home folder. No sudo, ever.
One command installs a LaunchAgent that runs the plan-and-apply at 04:30 with low I/O priority and an explicit PATH, logging what it removed and what it kept to ~/Library/Logs/mac-agent-hygiene.log.
Read the plugin on GitHub The full story and guide
Search, read and send WhatsApp messages from Claude through a local MCP bridge — the rail our client-group relays run on.
Move deliverables to Drive, keep durable memory in Notion, and snapshot Claude sessions so a long run can be resumed.
Governance for agent fleets: supervised orchestration patterns and an audit of where tokens go.
Everyday repair tools for the machine the agents live on: interactive disk exploration, the Chrome-extension bridge, stuck Cowork sessions.
Every plugin ships with install instructions; plugins declare their SOSA level in their manifest. Issues and pull requests are welcome on the repository.
Yes. Every plugin in the msapps-plugins marketplace is open source under the MIT licence and free to install in Claude Code or Cowork. Some plugins connect to third-party services that have their own accounts and pricing.
Only things that regenerate or already live somewhere else: Xcode DerivedData when no build is running, leftover plugin-install clones older than a day, the cache folders inside Playwright browser profiles, re-downloadable model caches, and git worktrees that are untouched for a week, clean, not in use and already pushed or merged. Logins, dirty or unpushed work and anything outside your home folder are never touched.
Yes. The plan step is a dry run that lists every candidate with remove or keep and a reason. Apply requires that plan's id and an explicit confirm, and re-checks each item at apply time. The scheduled daily run uses the same rules.
Yes. Each plugin declares its platforms; the marketplace is installable from Cowork via Settings → Plugins → Marketplaces and from Claude Code via /plugin marketplace add MSApps-Mobile/claude-plugins.
SOSA stands for Supervised, Orchestrated, Secured Agents — the MSApps methodology for running AI agents in production safely. Plugins declare their SOSA compliance level and impact classification in their manifest, and the methodology is published in the repository.
We build custom Model Context Protocol servers and Claude plugins that connect the model to your CRM, warehouse and internal tools — designed by a Claude Certified Architect. Preparing for a Claude certification yourself? Try the Cert Trainer by OpsAgents AI, an independent study tool for the Claude Certified exams, not affiliated with Anthropic.
Email michal@claudeservices.ai About our Claude practice