CONNECT
Pick your agent. Feint does the rest.
One install, three tools: attack, findings and fix. Every client that speaks MCP can run them.
Using an agent that’s not listed? Follow the manual setup guide and point it at the server below.
npx -y @feint/mcp
IN YOUR AGENT
Work from your agent. Stay in control.
Scan, triage and fix without leaving the editor
Install the Feint MCP server once. Ask for a scan, pull up a finding or open the fix PR from your chat. Scopes default to read.
Review every fix before it merges
Feint attacks your branch in an isolated copy and reports only what it broke. Claude Code writes the fix with Feint’s regression test; you review and merge.
ATTACK SURFACE
Six ways an agent gets owned.
Feint tries all of them.
Scanners see HTTP. They do not see a model deciding which tool to call. Feint does, multi-turn, inside an isolated copy of your stack.
01
Tool poisoning
A tampered tool description quietly rewrites what the model believes the tool does.
02
Indirect prompt injection
Instructions hidden in a web page, ticket or email arrive through a tool result and get followed.
03
Over-permissioned tools
One tool with write scope where read was enough. Feint finds the one that moves money or data.
04
Confused deputy
The agent uses its own credentials on behalf of the wrong tenant or user.
05
Secrets in tool config
Tokens in MCP configs, env files and transcripts. Found, chained, proven.
06
Runaway tool loops
Agents that keep calling tools under adversarial input, burning budget or taking irreversible actions.
Bring
Feint
to your AI product
Building an agent, an IDE or a platform? Connect it to the Feint MCP server so your users can scan, triage and fix without leaving it.
© 2026 Feint, Inc.
SOC 2 in progress · HIPAA · GDPR · Made in Framer