Choose a server you trust

Start with a server from a publisher you can identify. Read its source when available, recent release history, installation command, and requested permissions.

Ask four questions: What data can it read? What can it change? Where does it send data? Which credentials does it receive?

Also inspect the supply chain. A command such as npx -y package-name can download and execute the latest matching package under your user account. Pin a reviewed version when reproducibility matters, verify the actual registry owner and repository, and treat install scripts as executable code.

Permissions include more than configured folders. The process may inherit environment variables, network access, the current working directory, and operating-system privileges. A read-only tool schema does not prove the server process itself cannot write files.

For this exercise, choose a read-only server with a small scope. A local filesystem server limited to a disposable practice folder is easier to review than an account-wide administration server.

Create a temporary folder with two harmless text files. Do not point the exercise at your home directory, SSH keys, password stores, or real customer data.

Write a small trust record:

• exact package and version
• publisher and source repository
• command that runs
• filesystem and network scope
• credentials supplied
• capabilities advertised
• update and removal plan

Then run the server in the smallest practical operating-system and host scope. If a server requires broad access for a narrow task, choose another implementation or do not install it.

Repeat the review after upgrades. Trust applies to the code version you examined, not permanently to a package name.