If you use more than one AI coding tool, your MCP setup can become hard to keep track of.
You might add a server to Claude Code for one project, configure another in Cursor, and later import part of your setup into Codex. Each client has its own configuration, so it’s easy to forget what’s enabled where.
Here’s a simple review I find useful.
1. Make an inventory
Start by listing the AI clients you actually use and checking their MCP server settings. For each server, ask:
- Do I recognize it, and where did it come from?
- Which clients have it enabled?
- Do I still need it?
- What command does it run, or which remote endpoint does it connect to?
- What credentials and access does it need?
You can inspect each client’s settings manually. The goal isn’t to audit everything at once—it’s to know what you’ve connected.
2. Review after an import
When you move part of your setup to another client, check both the original and the new configuration. An import can leave you with servers enabled in places you no longer expect.
Pay particular attention to environment variables, authentication, headers and transport settings. A server that works in one client may need different settings in another.
3. Be careful with unfamiliar servers
A local MCP server is a program. Running a health check can start that program, so don’t health-check an unfamiliar server just to find out what it is.
Review the configured command and its source first. If you don’t recognize or need it, consider disabling or removing it from the relevant client.
4. Keep the setup manageable
Disable servers you don’t need in a particular client or workflow. Before editing configuration files, keep a backup so you can restore a working setup if something goes wrong.
A smaller, understandable setup is easier to review when something changes.
Where KyttoMCP fits
I build KyttoMCP, a free, open-source desktop app for macOS and Windows. It reads existing MCP configurations for Claude Desktop, Claude Code, Cursor, VS Code and Codex and shows servers across supported clients in one place.
You can review which clients have a server enabled, make changes to their existing configuration files and use backups when you need to go back. Kytto also offers health checks for servers you trust.
It’s a way to make your setup easier to inspect—not a guarantee that a server is safe. Review the source and permissions of the servers you use, and keep sensitive credentials out of anything you share publicly.
How do you keep track of your MCP servers when you use more than one AI client?
Top comments (0)