Back To Posts
Teaser (simple): MCP server security check guide
Sanket Joshi · November 17, 2026

AI assistants can now plug into other apps and tools.

That's powerful. It's also a little like installing a browser extension:
you're trusting code someone else wrote, with access you gave it.

Here's the part most people miss.

Each add-on tells the AI what it can do, in plain words.
And the AI reads those words like instructions.

So a badly written, or badly meant, add-on can quietly nudge your AI into doing things you never asked for.

I tested this myself.
I made a harmless-looking "add two numbers" tool that secretly asked the AI to hand over a private key.

A free scanner caught it in seconds.
But scanners only catch what they already know about.

My simple rule now, before plugging anything in:

✅ Check who made it
✅ Read what it says it does
✅ Give it the least access it needs
✅ Check again after every update

Scanners catch the obvious.
Judgment catches the rest.

I wrote a step-by-step guide for developers who want the full checklist. Link in the first comment.

Do you check your AI add-ons before you trust them? 👇

#AI #CyberSecurity #AIAgents

Originally posted on LinkedIn. Join the discussion there ↗