An MCP defense server prototype for detecting and sanitizing hidden-payload delivery to LLMs in agentic IDEs (e.g., Cursor). Built exclusively for ML security research and defensive hardening - no malicious use.
The MCP server is hosted on the cloud thanks to FastMCP Cloud. You can view the current deployment of the server at https://fastmcp.cloud/ronin-mcp/ronin-defense-mcp-v1/deployments. Whenever you commit and push a change to the main branch, it will automatically build and deploy onto the cloud. Please note that this link is only accessible to the owners of this project.
Go to File -> Preferences -> Cursor Settings -> Tools & MCP -> Installed MCP Servers. Create a new MCP server configuration using the following json:
{
"mcpServers": {
"ronin-defense-mcp-v1": {
"url": "https://ronin-defense-mcp-v1.fastmcp.app/mcp",
"headers": {}
}
}
}
Then, go back to Installed MCP Servers and refresh ronin-defense-mcp-v1 until the light is green and it shows the tools associated with the Ronin MCP server. If you had originally configured and turned on ronin-mcp-v1, please make sure that one is turned off in Cursor's Installed MCP Servers page because ronin-defense-mcp-v1 is a proxy for that malicious server. In Cursor, open the AI Panel and try the following prompt to see if the server connection is successful:
