Vulnify

Vulnify is a runtime authorization layer for AI agents. Before an agent reads, writes, deletes, exports or sends data, it describes the action and Vulnify returns ALLOW, REVIEW or BLOCK based on your organization's policies, a risk score and a sensitive-data scan. A REVIEW waits for a person on your team to approve or deny it in Vulnify or Slack.

This connector lets Claude ask those questions on your behalf:

• Check action: dry-run one action against your stored policies. Records nothing.
• Decide action: record a real decision. It writes a security event and an audit entry and may open a human review.
• Get decision: read a recorded decision, optionally waiting up to 30 seconds for a pending review to resolve.
• List policies: read your policies as code.
• Test policies: evaluate up to 200 test cases against stored or proposed policies without saving anything.

What it cannot do, on purpose: there is no approve or deny tool, so a REVIEW can only be resolved by a person in Vulnify. It cannot create, change or delete policies, and it does not run or forward the actions it checks.

You sign in with your Vulnify account (OAuth 2.1). The consent screen lists each permission, and you can revoke access at any time on the Connected apps page in Vulnify. Optional text you send for scanning is processed in memory and not stored.

Requires a Vulnify account (vulnify.io). Docs: https://docs.vulnify.io/mcp/