MCProbe · Know what an MCP server does before you let it in

Know what an MCP server does before you let it in.

Paste a server URL. MCProbe connects, reads every tool, safely tries them, and tells you in plain words whether it is safe to connect your AI agent.

An MCP server is a plug-in that gives an AI agent tools, such as search, files or email. Agents follow whatever text the server sends, so a bad server can steer them.

Try a deliberately malicious server, a safe one, or a real one →

Nothing gets installed. Tools are only called behind a safety gate that refuses anything destructive.

How it works

Four steps, about half a minute, nothing installed on your machine.

🔌

Connect

Opens a real MCP session over Streamable HTTP, with SSE as a fallback.

🔍

Inspect

Reads every tool, prompt and resource. Flags hidden instructions, bloated descriptions and dangerous capabilities.

🧪

Probe safely

An AI agent tries read-only tools with harmless inputs. A code-level gate blocks anything destructive.

🏁

Grade

Every finding is scored and you get a grade from A to F with the evidence behind it.

Recent scans

Every grade comes from a real run on this instance. Click one to open its report.