See which of the servers you run have been measured, and what was found. Nothing you paste leaves this page.
That is not a promise, it is a rule your browser enforces: this page is served with connect-src 'none', so no request can be made from it at all. The measurement index and the parser are built into the page, so once it has loaded it fetches nothing — open your Network tab and watch it stay empty while you type. The one request it makes is the webfont stylesheet, at load, before you have pasted anything.
Your env values are discarded as the config is read, before anything is drawn. The page lists which credentials each server asks for, by name, because that is the interesting part; the values never exist past the first step.
What it can and cannot tell you #
- A match is on package name . If you run a fork, or pin a version other than the one measured, the row says so.
- A remote server (
url, or typehttp/sse) is not measurable at Level 1: there is nothing to install and start. - A Python, Docker, Deno or Go server is not measured here either. Level 1 measures npm packages.
- No page is not a verdict. A server with no page here may never have been measured, or may have been measured with its page not yet published — a page reporting a finding goes up only once it has been re-checked by hand. The two look the same from here, deliberately: telling them apart would announce which packages have a finding before the people who ship them have seen the page. A maintainer who wants to know can ask: emailcontact@brydge-ai.com .
- Every row that links to a page carries that page's limits: one run, once, no tool called, not calibration-backed.
About the total #
It is a floor, not a figure. It adds up only the servers that have a page here, and it says above exactly how many of yours it left out and why. A server with no page, one hosted somewhere else, and one written in Python all count for nothing in the sum — so your real tool count and your real startup time are higher, by an amount this page has no way to work out.
The time is a sum of single runs. Each server's figure was measured once, on one machine, under a kernel tracer that stops the process at every file it opens, which makes it slower than the same start untraced. Adding such figures together compounds that, and nothing here corrects for it. Treat the total as an order of magnitude, not a measurement of your machine — your own startup depends on your hardware, your disk cache and which of these servers you start at once.
The card carries numbers and nothing else. No server name, no package name, no version, no credential name, no path. Not because those are secrets exactly, but because a list of what you run tells a reader more about your work than you probably mean to share along with a number.
[All measurements](https://mcp.brydge-ai.com/)
[How we measure](https://mcp.brydge-ai.com/method.html)
[Paste your config](https://mcp.brydge-ai.com/paste.html)
[Request a measurement](https://mcp.brydge-ai.com/request.html)
[Get an alert](https://mcp.brydge-ai.com/alerts.html)
[Pricing](https://mcp.brydge-ai.com/pricing.html)
[Harness and raw evidence](https://github.com/contactkontey-dotcom/brydge-mcp-measurements)
[BRYDGE’s calibrated measurements](https://www.brydge-ai.com/mcp)
Something on this page wrong? Email contact@brydge-ai.com, or open an issue. Corrections are free, and the correction is recorded.