Quick question for the builders running MCP servers:

How are you securing them, and where are you hosting them?

I’m concerned about the risks of connecting an LLM directly to my external tools.

Specifically:

Deployment: Are you running the MCP servers locally on your machine, inside Docker containers, or as a managed web service?
Isolation: How are you sandboxing them to prevent prompt injection from triggering dangerous actions?
Access: Do you use strict scope limits on API keys?

I want to make sure I’m not opening myself up to data leaks or unauthorized access before I scale this up.

5 comments

Quick question for the builders running MCP servers: