Category: Uncategorised

amtiYo/agents: One .agents source of truth to sync MCP servers, skills, and instructions across Codex, Claude Code, Gemini CLI, Cursor, Copilot, and Antigravity.


npm version
License: MIT
Tests

One config to rule them all.
The practical standard layer for multi-LLM development.

You’re using Codex, Claude, Cursor, and Gemini. You have 6 different config files. You add an MCP server to one, and the others break. Your team’s onboarding doc is a wiki page that’s always out of date.

agents gives you one source of truth.

Define your configuration once in .agents/, and let the CLI handle the rest. We sync your MCP servers, skills, and instructions to every tool automatically.

.agents/
  ├── agents.json      → All your MCP servers
  ├── local.json       → Your secrets (gitignored)
  └── skills/          → Reusable team workflows

npm install -g @agents-dev/cli

Go to your project folder and run:

The interactive wizard will guide you through:

  • ✅ Creating the .agents folder
  • ✅ Detecting your installed AI tools
  • ✅ Setting up your first MCP server

Whenever you change your config, just run:

Boom. All your tools are updated.

Tool MCP Integration Skills Instructions
Codex ✅ ✅ ✅
Claude Code ✅ ✅ ✅
Gemini CLI ✅ ✅ ✅
Cursor ✅ ✅ ✅
Copilot ✅ ⏳ ✅
Antigravity ✅ ✅ ✅

Command Description
agents start Start here! Interactive setup wizard.
agents sync Syncs your .agents config to all tools.
agents mcp add Adds a new MCP server (e.g., from a URL).
agents mcp list Lists all configured MCP servers.
agents doctor Checks for configuration issues.
agents status Shows connection status of all tools.
agents watch Auto-syncs changes in real-time.

  1. You edit .agents/agents.json (or use the CLI).
  2. agents reads your standard configuration.
  3. agents generates the specific config files for each tool (e.g., .cursor/mcp.json, .claude/mcp.json).
  4. You focus on coding, not configuring.
Does this replace AGENTS.md?

**No.** It *extends* it. `AGENTS.md` is for human-readable instructions. `agents` is for machine-readable configuration (MCP servers, skills). We support both.

Can I use this with just one tool?

**Yes!** It’s still better than managing raw config files because it gives you a clean git strategy (commit `.agents`, ignore generated files) and easy MCP management commands.

Where are my API keys stored?

In `.agents/local.json`. This file is **gitignored** by default, so your secrets stay safe on your machine.

Built with ❤️ for the AI coding community



Source link