Skip to main content
PromptGuard works in Cursor three ways, which you can combine: the MCP server gives the Cursor agent PromptGuard tools, the editor extension marks unprotected LLM SDK calls in your code, and the PromptGuard plugin’s rules and commands steer the agent toward secure LLM code.

​
MCP server

Open this link in Cursor to add PromptGuard’s hosted server, then click Needs login next to it and sign in with your PromptGuard account. There is no API key to copy:
It adds this entry to your Cursor MCP configuration:
The agent can then scan text for prompt injection and other threats, redact PII, and find LLM SDK calls in code you give it. To let it scan the files of your project instead, add the local server, which runs the PromptGuard CLI on your machine. The tools each server offers, and their parameters, are on the MCP server page.

​
Editor extension

The PromptGuard extension runs in Cursor, but it is not on Open VSX, so Cursor’s extension search does not find it. Download promptguard-vscode-<version>.vsix from the latest release, then run Extensions: Install from VSIX… from the Command Palette and choose the file. It is not updated automatically: install each new release’s .vsix the same way.

​
Rules and commands

The PromptGuard plugin repository holds Cursor rules and agent commands. It is not listed in the Cursor Marketplace; to use them, copy them into your project:
  • rules/secure-llm-usage.mdc and rules/llm-security-review.mdc into .cursor/rules/
  • commands/promptguard-scan.md and commands/promptguard-secure.md into .cursor/commands/
The same repository packages skills and a security-reviewer agent for Claude Code and Codex; its README covers those.
Last modified on November 4, 2026