Your Claude Code, Cursor, and Gemini CLI are cluttered with MCP Servers, but you have no idea which ones are working and which aren't? This guide will help you sort it all out.
What is MCP?
MCP (Model Context Protocol) is an open standard introduced by Anthropic in late 2024. It defines how AI models connect to external tools, databases, and services through a unified JSON-RPC 2.0 interface.
In simple terms: MCP is the "USB-C port" for AI coding tools.
Before MCP, each AI tool had its own plugin system. Cursor used one way to connect external services, Claude Desktop used another, and Claude Code used a third. MCP unifies everything.
The MCP Server Explosion Problem
As MCP adoption grows, a new problem emerges: MCP server sprawl.
Common Symptoms
- You've installed 20+ MCP servers across different tools
- Some stopped working months ago but you never cleaned them up
- You're not sure which tool has which servers configured
- Multiple copies of the same configuration exist in different config files
- Legacy Node.js-based MCP servers clutter your config alongside Python ones
Why This Happens
Each MCP-compatible tool maintains its own configuration:
-
Claude Code:
~/.claude/mcp.json -
Claude Desktop:
~/Library/Application Support/Claude/claude_desktop_config.json - Cursor: Settings → MCP
- VS Code: Via extensions
- Gemini CLI: Separate config
There's no centralized management—each tool has its own config format and location.
How to Audit Your MCP Servers
Step 1: Inventory What You Have
# Claude Code
cat ~/.claude/mcp.json
# Claude Desktop (macOS)
cat ~/Library/Application\ Support/Claude/claude_desktop_config.json
Step 2: Test Each Server
# Test if an MCP server is working
echo '{"jsonrpc":"2.0","id":1,"method":"ping"}' | nc localhost PORT
Step 3: Categorize
| Status | Action |
|---|---|
| Working and used | Keep |
| Working but unused | Remove |
| Broken | Fix or remove |
| Duplicate | Consolidate |
Best Practices for MCP Management
1. Standardize on Python MCP Servers
Newer MCP servers use the Python SDK. If you have legacy Node.js servers still running, consider migrating them to Python for consistency.
2. Use Environment Variables
Instead of hardcoding paths and API keys, use environment variables referenced in your MCP configs:
{
"mcpServers": {
"custom-server": {
"command": "python",
"args": ["-m", "my_server"],
"env": {
"API_KEY": "${MY_API_KEY}"
}
}
}
}
3. Document Your Setup
Keep a record of which MCP servers you have, what they do, and which tools they're configured for. When you debug issues, this documentation is invaluable.
4. Regular Cleanup
Set a monthly reminder to audit your MCP servers. Remove unused ones, update broken configurations, and consolidate duplicates.
5. Centralized Management with Skills
If you use Claude Code extensively, consider centralizing MCP configurations as re-usable skills. Skills Hub in Mantra allows you to organize and version your configurations alongside your prompts and workflows.
Troubleshooting Common Issues
Server Won't Start
Check:
- Is the command path correct?
- Are all dependencies installed?
- Is the Python/Node.js version compatible?
Server Starts but No Response
Check:
- Is the transport protocol correct (stdio vs SSE)?
- Are there port conflicts?
- Check logs for errors
Authentication Failures
Check:
- Are API keys valid?
- Are environment variables properly set?
- Is the authentication scope correct?
The Future of MCP Management
As the MCP ecosystem grows, expect better tooling for server management. The protocol itself is evolving, with improvements in discovery, authentication, and monitoring on the roadmap.
For now, the key is staying organized: know what you have, keep it clean, and document your setup.
Mantra's Skills Hub supports organizing MCP configurations alongside your prompts and workflows. Learn more at mantra.gonewx.com.
Top comments (0)