Vulnerabilities in Anthropic’s MCP: Full-Schema Poisoning + Secret-Leaking Tool Attacks (PoC Inside)

New research has been published, revealing critical vulnerabilities in Anthropic's Model Context Protocol (MCP). The findings include Full-Schema Poisoning attacks that inject malicious logic into any schema field and advanced Tool Poisoning techniques that induce LLMs to disclose secrets such as SSH keys. These stealthy attacks only trigger in production.