Observed arrival · 2026-10-12
MCProof puts MCP agents through measured attacks
MCProof runs attacks against an MCP-connected AI agent, records its tool calls, and reports whether it was compromised.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- MCP agent developers and security testers
- Worth noticing
- The homepage reports 0.96 precision and 1.00 recall across 51 labeled realizations, but says those figures are valid only for the named frozen judge.
Field notes
The page shows an eight-step sample replay in which a tool-misuse attack reaches an out-of-scope file read, and links to additional attack-category samples. Its displayed detection figures include sample sizes, a named judge, and an explicit warning that changes to the rubric, model, or temperature require fresh measurement.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
⊠LoginAccess appeared gated
✦PrettyNotable craft visible
◎NicheUnusually specific use
ƒJavaScriptBrowser-side code central
One card from the complete issue