Lesson 163 · 3 min
MCP safety: what can go wrong with connectors
Every connector is a door between the AI and your data, so check who made it, what it can touch and what text it lets in.
An MCP server runs code and handles your data. A well-known company's official server is one thing. A random server copied from a forum is another, and it could send your files somewhere else without telling you.
The second risk is what flows in. If a connector reads emails, web pages or shared documents, any of them could contain hidden instructions aimed at the AI. Mixing a tool that reads untrusted text with a tool that can send or delete is where most trouble starts.
The third risk is too much power. A server that can read, write, delete and send money may be convenient, but one wrong call or one tricked model can do a lot of damage at once.
Use official or well-reviewed servers, give read-only access where you can, turn on confirmation for any action that changes things, and look at the activity log from time to time. In a company, let IT keep an approved list.
Do it now
Audit your connectors
- 1. List every connector, plugin or extension your AI tools use.
- 2. For each one, write who made it and whether it can read only or also change things.
- 3. Remove one you do not need and switch one to read-only if possible.
Quick check
Keep going
Next →Why use more than one AI togetherWords from this lesson
- MCP
- An open standard that lets AI assistants plug into apps and data sources, like a universal adapter for tools.
- Prompt injection
- Hidden instructions planted in a web page, email or file that try to hijack an AI reading it.
- Data privacy
- Keeping personal or confidential information under control, including deciding what you type into AI tools and where it is stored.
- Guardrails
- Rules and filters around an AI system that block harmful, off-topic or unsafe requests and answers.