root@mcp-hub:~$
mcp-hub.init

MCP//HUB

> init trust-layer...
[  OK  ] catalogue ............. 13178 servers indexed
[  OK  ] behavioral pentest .... isolated micro-VM (Firecracker)
[  OK  ] monitoring ............ 24/7 (rug-pull detection)
> ready

The MCP registry that actually tests servers.

MCP Hub continuously audits the whole MCP ecosystem: token cost, behavioral pentest in an isolated sandbox, and change monitoring. Know whether a server is safe before you connect it to your AI agent.

Browse verified serversCreator space — get your server verifiedBuild a safe serverHow it works
13,178servers in the directory
2 layersdescription + real pentest
24/7change monitoring
Firecrackereach tool in a micro-VM
three pillars, one dataset

Benchmarks

Context cost (tokens to load all tools) and tool-selection quality. No other registry measures it.

Two-layer security

Description analysis and a behavioral pentest: we actually call the tool and watch. A description can lie; behavior can't.

Continuous monitoring

The whole registry re-scanned continuously. We catch silent changes (rug-pulls) and flag them.

how we actually test security

Scanners stop at the description, in a CLI, client-side. We go all the way to real behavior: each server runs in an isolated Firecracker micro-VM with decoys and blocked, logged networking. Reading a secret or exfiltrating data is caught on the spot. Only this behavioral test yields a verdict — a description alone is never treated as proof.