I've been building AI agents in LangChain, n8n, and some custom Python scripts. Each time I need to connect tools (Slack, GitHub, etc.), I'm recreating the same MCP server configs, dealing with auth, and managing dependencies separately for each agent. I'm curious how others are handling this: - Are you running MCP servers separately for each agent or sharing them? - How do you manage credentials and configs across frameworks? - Any issues with dependency conflicts (like Agent A needs Python 3.9 but Agent B needs 3.11)? - Do you have observability into which agent is calling which tool? I started building a private registry to centralize this, but wondering if I'm overengineering or if others hit the same pain points. What's your setup look like?