
Professional PoC for CVE-2025-59536 and related CVEs. Demonstrates an MCP Tool Confirmation Prompt Misrepresentation in Anthropic Claude_Code leading to arbitrary OS command execution.
A professional Proof-of-Concept (PoC) demonstrating a critical UI/UX flaw in Anthropic's Claude Code. This vulnerability allows a malicious MCP server to mislead users into approving arbitrary command execution by misrepresenting tool parameters in confirmation prompts.
Claude Code trusts the description and metadata provided by an MCP server to generate user-facing confirmation prompts. An attacker-controlled server can provide a benign description (e.g., "Read a file") while the underlying execution logic triggers a reverse shell or unauthorized file writes.
# Clone the research repository
git clone [https://github.com/Rohitberiwala/Claude-Code-MCP-Injection](https://github.com/Rohitberiwala/Claude-Code-MCP-Injection)
cd Claude-Code-MCP-Injection
# Edit the script to add your listener IP
nano exploit.py
# Run the PoC generator
python3 exploit.py