macOS · Native App
AI Agents Manager · macOS · Native App

Claude, Codex, Gemini, Cursor…
Instructions, Skills, MCP.
All in sync.

You use Claude Code, Cursor, Gemini, Windsurf and more — each with its own skills, instruction files, and MCP servers. The same configuration lives in 10+ directories. This app keeps all of them in sync.

Download Free See All Features
Claude Code Gemini CLI Cursor Windsurf Copilot Codex Cline Roo Code Continue.dev Augment Aider Amazon Q + 13 more
The Problem

Every AI coding agent stores its skills, instructions, and MCP servers in its own directory. The same information needs to exist in 10+ places at once — and there's no built-in way to keep them in sync.

Modern developers use multiple AI coding agents simultaneously — Claude Code, Gemini CLI, Cursor, Windsurf, Copilot, Codex, Cline, and more. Each has its own config directory, its own skills folder, its own instruction files (CLAUDE.md, .cursorrules, .windsurfrules), and its own MCP server config.

You add a skill to Claude Code — Cursor doesn't have it. You configure an MCP server in ~/.claude.json — it doesn't exist in ~/.cursor/mcp.json. You update your .cursorrules — your CLAUDE.md still says the old thing. Every agent is its own island.

And there's a performance problem: loading too many skills into an agent's context wastes tokens, causes false triggers, and slows responses — but without tooling, there's no easy way to control which skills are active and which aren't.

  • ⚡ Skills scattered everywhere Duplicates across agent directories. Version drift. No single source of truth.
  • 📄 Instructions diverge Each agent's .md file drifts out of sync, leading to inconsistent behavior.
  • 🔌 MCP servers configured individually Set up the same server in 10 different config files. Miss one and it's broken.
  • 🐌 Too many skills kill performance 50 skills loaded = thousands of wasted tokens before the agent reads your question.
  • 🔒 No safe way to experiment One bad edit can break an agent setup with no rollback.
How It Works

One shared directory for all your skills. Symlinks connect it to every agent on your machine — Claude Code, Cursor, Windsurf, Gemini, and 20+ more.

The app treats ~/.agents/available_skills/ as the single source of truth. Every agent's skills/ directory is a symlink pointing here. Add a skill once — every agent sees it. Update a skill once — every agent gets the update. No file copying. No version drift.

available_skills/

The Library — Everything You Own

Your complete collection of skills. Every skill you've ever installed lives here as a real folder. Think of it as your bookshelf: every book you own, whether you're currently reading it or not. Skills accumulate here and are never deleted during normal operations.

skills/ (symlinks only)

The Active Set — What Agents See

Contains only symlinks pointing back to available_skills/. Each symlink means "this skill is currently active." Agents read only from here — so only the skills you've explicitly activated are visible. You might have 80 in the library, but only 10 loaded.

~/.agents/ ├── available_skills/ ← The Library: ALL skills you own (real folders) │ ├── skill-a/ │ ├── skill-b/ │ ├── skill-c/ │ └── .library_meta.json ← tracks source + install date per skill ├── skills/ ← The Active Set: ONLY what agents see (symlinks) │ ├── skill-a -> ../available_skills/skill-a │ └── skill-c -> ../available_skills/skill-c │ (skill-b is available but NOT active) ├── agent.md ← Master instructions (all agents read this) └── backup/ ← Timestamped backups of everything ~/.claude/skills/ -> ../../.agents/skills/ (symlink) ~/.cursor/skills/ -> ../../.agents/skills/ (symlink) ...every other agent...
Features

Everything you need to manage a multi-agent setup — from a single dashboard to per-project skill activation.

Feature 01
Status Dashboard

A single-screen health check of your entire multi-agent setup. One glance tells you if anything is out of sync across all 25+ agents.

  • Per-agent status: backup, consolidation, symlinks, instructions, MCP sync
  • Summary KPIs: active skills, merged instructions, MCP servers, launcher status
  • No wildcard filesystem scanning — no macOS privacy prompts
  • Discovers 25+ known agent paths automatically
AI Agents Manager Status Dashboard
Status Dashboard
Feature 02
Active Skills Toggle

A checkbox list of every skill in your library with one-click activate/deactivate. Keep 5–15 skills active. Keep the rest available for when you need them.

  • Filter tabs: All / Active / Inactive with counts
  • Search bar with instant filtering
  • Source filter: see where each skill came from
  • Toggle creates/removes a symlink — agents see the change immediately
  • Select All / Deselect All for bulk operations
Active Skills Toggle View
Active Skills
Feature 03
Skills Library

An App Store-like experience for AI agent skills, backed by GitHub repositories. Your team publishes skills — every developer's manager shows new and updated skills in one click.

  • Connect any GitHub repo as a skill source (public or private)
  • Status per skill: Not Installed / Up to Date / Update Available
  • Filter tabs: All / New / Updates / Installed
  • Install downloads all files recursively including subfolders
  • "Update All" button when multiple updates exist
Skills Library with GitHub sources
Skills Library
Feature 04
Manage Sources

Configure one or more GitHub repositories as skill sources. Supports full repo paths, subdirectories, and single-skill URLs. Private repos with token support.

  • Multiple source formats: owner/repo, owner/repo/path, full GitHub URLs
  • Optional GitHub token for private repositories
  • Enable/disable toggle per source
  • Uses GitHub Tree API — single call enumerates all files
Manage Skill Sources
Manage Sources
Feature 05
Instructions Sync

Consolidates all agent-specific instruction files into one master document. Write your instructions once — every AI agent reads the same guidelines.

  • Supports 14+ instruction file formats (CLAUDE.md, .cursorrules, .windsurfrules, .clinerules, AGENTS.md…)
  • Backup → Merge → Redirect in one operation
  • Per-file editor with redirect toggle
  • Footer enforcement: ensures agents always know about the skills system
  • Generate missing instruction files for agents that don't have one yet
Instruction Sync and Redirect
Instructions Sync
Feature 06
MCP Server Sync BETA

A unified view and sync tool for Model Context Protocol server configurations across all agents. Configure once, sync everywhere.

  • Consolidated matrix: every server × every agent at a glance
  • Supports Claude Code, Claude Desktop, Gemini CLI, Cursor, Windsurf, VS Code, Codex
  • "Sync to All" — adds missing servers, never overwrites existing entries
  • Toggle enabled/disabled per agent, delete globally with backup
  • Timestamped backup & restore for all MCP config files
MCP Server Configuration Matrix
MCP Sync — BETA
Feature 07
CLI Launchers

Shell wrapper scripts that intercept bare claude or gemini commands and present a quick menu — so you never type claude --continue --dangerously-skip-permissions again.

  • Claude menu: Continue, Resume, New Chat — all with permission options
  • Gemini menu: Resume latest, Resume, New chat — yolo or normal mode
  • Arguments still pass through transparently for scripted usage
  • One-click install/uninstall — adds PATH entry to .zshrc automatically
CLI Launcher Menu
CLI Launchers
Team Skills

How do you make sure every developer on your team uses the same skills, follows the same AI instructions, and has access to the same tools — without anyone manually copying files?

In most teams, AI skills are ad hoc. One developer has a lead-qualifier skill that took hours to refine. Another has a code-review skill no one else knows exists. A third added a CLAUDE.md that contradicts what everyone else has.

Skills are institutional knowledge. When they live on one machine, that knowledge is invisible to the rest of the team. The AI Agents Manager solves this with a GitHub-backed team skills library — a private or public repo that becomes the single source of truth for every skill your team uses.

1
repo for all team skills
0
manual file copies needed
∞
agents stay in sync
01

Create a private GitHub repo for your team's skills

Organise skills as folders in the repo — one folder per skill, just like ~/.agents/available_skills/ on your machine. Each skill has its own SKILL.md, scripts, and assets. The repo is the canonical version of every skill your team owns.

# repo structure
your-org/ai-skills/
  ├── lead-qualifier/
  ├── code-reviewer/
  ├── sprint-planner/
  └── onboarding-guide/
02

Add it as a source in the Skills Library

In the Skills Library view, click Add Source and paste your repo slug — your-org/ai-skills or the full path to a subfolder. For private repos, add a GitHub personal access token with repo:read scope. The token is stored locally and never leaves your machine.

Source your-org/ai-skills Private
Token github_pat_•••••••••••• Connected
03

Every developer installs from the same source

The Skills Library shows each skill's install status: Not Installed, Up to Date, or Update Available. New team members open the app, add the repo as a source, and install everything with one click. No Slack threads. No shared drives. No "which version do you have?"

lead-qualifier Up to Date
code-reviewer Update Available
sprint-planner Not Installed
onboarding-guide Up to Date
04

Update once — everyone gets it

When you improve a skill, push the changes to the repo. Every team member's Skills Library shows an Update Available badge. One click on Update All and they're in sync. No version drift. No "I thought you updated that". The SHA of every installed skill is tracked locally so the app knows exactly what changed.

🔒

Private repos with token auth

Add a GitHub personal access token for private team repositories. Tokens are stored locally in the app — never sent to any server.

🌿

Multiple sources

Add several repos as sources — an official company library, a personal collection, and an open-source community pack. Each skill shows which source it came from.

📦

Full folder install

Skills are downloaded recursively — all files, subfolders, scripts, and assets. The same structure as the repo lands in ~/.agents/available_skills/.

🔁

SHA-based update detection

The app compares the local tree SHA against the remote. If a single file changed, the skill is marked as needing an update. No false positives. No missed updates.

More Features
Feature 08

Skills Sync

Backup → Consolidate → Symlink in one operation. Moves all skills from every agent into the shared store, resolves conflicts via a UI modal (keep / replace / keep both), restores active links automatically.

Feature 09

Per-Project Skills

Activate different skill subsets per project directory. A web project gets frontend-design while an infra project gets terraform-helper. Each agent sees exactly what it needs.

Feature 10

Global Instructions Editor

In-app editor for ~/.agents/agent.md. Cmd+S shortcut, modification tracking, footer enforcement badge — green checkmark if the mandatory footer is present, auto-appended on save.

Feature 11

Auto-Update System

Checks for new versions every 4 hours from a Railway-hosted endpoint. Downloads, shows a progress bar, replaces the app bundle, and relaunches. Dismissable per version — non-intrusive.

Feature 12

Logs Viewer

Color-coded log viewer for all sync operations. Levels: Header, Info, Success (green), Warning (yellow), Error (red). After any sync, a "View Logs" link shows the full output for debugging.

Always Safe

Backup First

Every destructive operation — Skills Sync, Instructions Sync, MCP Sync — creates a timestamped backup first. Conflict resolution is explicit, never silent. No safe way to accidentally break your setup.

Native macOS App · Free · 25+ Agents Supported
One app for all
your AI agents.

Skills, instructions, and MCP servers — managed in one place, synced to every agent on your machine. Install once. Everything stays in sync.

Download for macOS
macOS 13 Ventura or later · Apple Silicon & Intel · Free
FAQ

Common questions about managing skills, instruction files, and MCP servers across multiple AI coding agents.

How do I manage skills across Claude Code, Cursor, Windsurf, and other agents?

The AI Agents Manager uses a two-folder system: ~/.agents/available_skills/ stores all your skills as real folders — your complete library. ~/.agents/skills/ contains only symlinks to the skills you want active right now. Each agent's own skills/ directory is symlinked here.

One click in the Active Skills view creates or removes a symlink. Every agent — Claude Code, Cursor, Windsurf, Cline, and more — sees the change immediately because they all read from the same shared directory. No file copying. No version drift.

How do I sync MCP servers across Claude Code, Cursor, and Windsurf?

The MCP Configs view shows a cross-reference matrix: every MCP server × every agent. Green checkmark = configured, dash = missing. One click on "Sync to All" adds missing servers to:

  • Claude Code — ~/.claude.json
  • Cursor — ~/.cursor/mcp.json
  • Windsurf — ~/.codeium/windsurf/mcp_config.json
  • Gemini CLI — ~/.gemini/settings.json
  • VS Code — ~/.vscode/mcp.json
  • Codex — ~/.codex/config.toml

It only adds — never overwrites or removes existing entries. Timestamped backups are created before any change.

What is the difference between cursor rules and windsurf rules?

Cursor reads agent instructions from .cursorrules or .cursor.md. Windsurf reads from .windsurfrules or .windsurf.md. Both tell the AI how to behave in your project — they're the same concept with different filenames.

The AI Agents Manager consolidates all 14+ instruction file formats (cursor rules, windsurf rules, CLAUDE.md, GEMINI.md, .clinerules, .roorules, AGENTS.md, and more) into a single ~/.agents/agent.md master document. Each agent's original file is then rewritten to redirect to the master — so you write instructions once and every agent follows them.

How do I set up Claude Code MCP servers?

Claude Code MCP servers are configured in ~/.claude.json under the mcpServers key. The AI Agents Manager reads this file, displays all your MCP servers in a visual matrix, and lets you sync any server configuration from Claude Code to every other agent in one click.

It supports both stdio and http transport types. You can also toggle individual servers enabled/disabled per agent, or delete a server globally (with automatic backup first).

What is the too-many-skills problem in AI coding agents?

When an AI coding agent starts, it scans its skills/ directory and loads all skill metadata into its system prompt. With 50 skills loaded, you're burning thousands of tokens on skill descriptions before the agent reads your first message. This causes:

  • Slower first responses and higher API costs
  • Skill confusion — wrong skills triggering on irrelevant tasks
  • Conflicting instructions from incompatible skills

The solution: keep your full collection in available_skills/ but expose only 5–15 active skills via symlinks in skills/. The Active Skills toggle makes this a one-click operation per skill.

Which AI coding agents does the manager support?

25+ agents are supported out of the box: Claude Code, Gemini CLI, Cursor, Windsurf, GitHub Copilot, Codex CLI, Cline, Roo Code, Continue.dev, Augment, Amazon Q, Tabnine, Aider, Junie (JetBrains), Devin, cagent, VS Code, Zed, and more — including sub-agents like Antigravity (nested under Gemini). New agents are added by editing a single array in the configuration.

Is the AI Agents Manager free?

Yes, the app is free to download and use. It is currently pending approval in the Apple Developer Program. In the meantime, you can install it directly via the install script: curl -fsSL https://makeitfuture.up.railway.app/install.sh | bash