I built a deterministic scanner that catches command injection, prompt-injection markers in tool descriptions, over-broad/destructive tools, and committed secrets in MCP servers before you connect them. It continuously scans the entire official MCP registry and makes results available at https://index.canopii.dev.
Now the engine behind that scanner is available as a CLI that you can integrate into your processes and CI/CD pipelines (e.g --min-grade B).
If you are building MCP servers, you can make sure your MCP server is secure using the local mode before publishing it. Or check why your server scored low, apply fixes, check if your score has improved locally first before publishing a new version so as to eliminate guesswork.
If you are a user, easily check the security of an MCP server before connecting your AI agents using one of the available methods (remote, github, pypi, npm etc.)
You can also use an LLM as a "judge". It won't affect the score but it may flag important security considerations that a deterministic scanner can't catch.
No telemetry, no account necessary.
Github: [https://github.com/canopii-dev/canopii-cli](https://github.com/canopii-dev/canopii-cli) (license: Apache 2.0)
Comments URL: [https://news.ycombinator.com/item?id=49094760](https://news.ycombinator.com/item?id=49094760)
Points: 1
source & further reading
index.canopii.dev — original article