How we grade
Every grade is calculated by a script from things we can measure. Nobody picks a grade by feel.
The five checks, 20 points each
1. Runs
We install the package in a throwaway folder with an empty environment and start it. If it completes the MCP handshake we give it points. If we know a safe tool to call, we call it, and the result counts too. A server that does not start gets an F.
2. Tool descriptions
An agent decides which tool to use from its description. We check how many tools have a real description and how many parameters are explained.
3. Package safety
We look at public package data: install scripts that run code on install, a very large dependency tree, a missing license, a missing source repository. This is a quick screen, not an audit.
4. Maintenance
How recently the package was published. A deprecated package scores near zero here.
5. Adoption
Weekly downloads from npm. More people using it means more eyes on problems.
Letter grades
A is 90 and up, B 80 to 89, C 70 to 79, D 60 to 69, F below 60 or if it does not start.
The badge
"Tested by ConnectATD" means the server started, answered, and has not been deprecated by its publisher. It does not mean we audited the code.
Limits we are upfront about
- We test npm packages first. Other kinds come later.
- Servers that need an account or key are tested for startup and tool listing only, never with real credentials.
- Scores change. Each page shows the test date and package version.
Spotted something wrong, or you maintain a server and want a retest? Tell us through dealithic.co.