Managing Multiple MCP Configurations for Different Projects

As a developer, you rarely work on just one project. You might have:

• A Work Project (Enterprise codebase, strictly secured).
• A Side Project (New tech stack, experimental tools).
• A Learning Repo (Tutorials, minimal setup).

Using the same MCP configuration for all of these is dangerous and inefficient.

The “One Giant Config” Problem

If you put every MCP server into your global claude_desktop_config.json, two things happen:

1. Context Pollution: You ask about your side project’s DB, but the AI accidentally queries your work DB because both servers are active.
2. Performance Hit: Launching 20 servers consumes RAM and CPU, even if you only need 3.

Strategy 1: Manual Swapping (The Hard Way)

You could maintain multiple JSON files (config.work.json, config.personal.json) and manually rename them before launching Claude.

• Pros: Free.
• Cons: You will forget. You will accidentally commit work secrets to personal repos.

Strategy 2: Project-Scoped Profiles (The Vibe Manager Way)

Vibe Manager introduces the concept of Profiles.

How It Works

1. Create a profile named “Work”.
   • Enable: Enterprise GitHub, PostgreSQL (Local), Jira.
2. Create a profile named “Creative”.
   • Enable: Brave Search, Image Generator, FileSystem (limited to ~/art).

Switching Contexts

Before you start your day, you select “Work Profile” in Vibe Manager and click Sync. Instantly, Claude and Cursor are reconfigured to only see the tools relevant to your work.

Best Practices for Multi-Project MCP

1. Restrict FileSystem Scope: Never give a global FileSystem server access to /. Always scope it to specific project paths.
2. Separate API Keys: If using cloud tools, ensure your personal profile uses your personal API key billing.
3. Naming Conventions: Prefix your server names if you must run them simultaneously (e.g., work-postgres vs personal-postgres).

By isolating your environments, you keep your AI focused and your data secure.