I built a CLI that unifies configuration for AI coding assistants.

Problem: Modern AI-powered development uses multiple tools (Cursor, Claude Desktop, GitHub Copilot, various CLI tools). Each has its own config file for MCP (Model Context Protocol) servers. Adding a server means manually updating 4-6 different configs.

Solution: agents gives you one source of truth. Define MCP servers once, sync everywhere.

Example: .agents/ ├── agents.json → Your config (committed) ├── local.json → Secrets (gitignored) $ agents sync → Updates all tool configs

Key decisions: - Convention over configuration (sensible defaults) - Security-first (secrets separated, auto-gitignored) - Interactive setup wizard for easy onboarding - 70+ tests (Vitest)

Supports: Cursor, Claude Code, Gemini CLI, Codex, Copilot, Antigravity

GitHub: https://github.com/amtiYo/agents npm: @agents-dev/cli

v0.7.7 released today. MIT licensed.

Looking for feedback on: 1. Is this solving a real problem? 2. What other tools should it support? 3. Thoughts on the architecture?

Happy to answer questions about the implementation or design decisions.