Protecting against tool poisoning attacks

MCP Defender protects against tool poisoning by including every detail of the tool call, such as the tool description, name, parameters, types into every scan it performs. They are all checked for malicious instructions intended to manipulate the agent. Here's a simple diagram of a tool poisoning attack both with and without MCP Defender.

If you find a way to circumvent the tool poisoning protection, please submit a PR and add it to the dataset! Successful attacks will receive a small award as a token of our appreciation. We'll continue to update our tool poisoning protections as new attacks are discovered.

Additional Reading

• https://www.cyberark.com/resources/threat-research-blog/poison-everywhere-no-output-from-your-mcp-server-is-safe

• https://invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks

Learn more about how MCP Defender secures Cursor and other AI apps