flankerhqd/cyvisguard

View on GitHub ↗

Security control plane for AI agents — identity and delegation, capability policy, data-flow taint and a live audit trail, enforced over MCP. Guards a real Claude Code end to end.

54 ★4 forksTypeScriptUpdated 1mo ago

What you need to know

Open-source, self-contained security control plane for AI agents. Sits between an agent and everything it reaches, answering who is calling, on whose behalf, and whether it is allowed, enforced over MCP. Ships a reference integration with Claude Code and guards tool calls, refuses prompt-injected exfiltration, and catches poisoned CLAUDE.md files, with a live audit console.

Key features

  • Security control plane placed between agents and their tools
  • Identity and authorization checks over MCP
  • Guards tool calls and refuses prompt-injected exfiltration
  • Detects poisoned CLAUDE.md / instruction files
  • Live audit console
  • Reference integration with Claude Code

Reviewed 2026-08-11

Topics

agent-securityai-agentsai-securityclaude-codellm-securitymcpmodel-context-protocolprompt-injectionzero-trust
Stars
54★
Forks
4
Language
TypeScript
License
Apache-2.0
Created
2026-07-22
Last push
2026-07-27