The gives AI coding agents real context about your project — here's how it works

#mcp #ai #programming

The Problem

If you use Claude Code, Cursor, or any AI coding assistant daily you've probably run into this:

The agent doesn't know your project. It knows your conversation.

Every session you're re-explaining the same stack, pasting the same package.json, describing the same environment. It's friction that adds up fast especially in a team setting where everyone has a slightly different setup.

Zephex is a CLI tool that connects your actual codebase to your AI agent through MCP (Model Context Protocol).

One command:

mcpcli setup

It opens a browser auth flow, you log in, and then it scans your project. Not just the name — everything. Packages, versions, environment variables, full stack detection.

What It Found in a Real Project

I ran it on a Next.js monorepo using Stripe, Supabase, and Auth0. Here's what it surfaced:

Package	Installed	Latest	Risk
stripe	20.4.1	22.2.1	Major gap / Medium
@supabase/ssr	0.3.0	0.12.0	Minor gap / Low
@supabase/supabase-js	2.106.1	2.108.2	Patch / Low
@sentry/nextjs	—	—	Patch / Low

It also found 27 environment variables across 6 categories and flagged a TypeScript version mismatch. No critical CVEs — but a clear table of what to look at next.

How the Editor Integration Works

After the scan, you link it to your editor. I used OpenCode in the demo, but it works with Claude Code, Cursor, VS Code, Kiro, Zed, and Windsurf.

The connection runs through MCP — so the AI agent inside your editor has that context at runtime. When you ask "should I update anything?" it gives you a real answer, not a generic one.