Launching today
QAmcp Monitor
Know when an AI can’t use your MCP server anymore
2 followers
Know when an AI can’t use your MCP server anymore
2 followers
QAmcp Monitor tests your MCP server the way an AI app uses it and emails you when it goes down, a tool changes, the score drops or a key shows up in an answer. Free for 1 server. Built on QAmcp, a free open-source MCP tester.



Free Options
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt, I’m Abhiram, a solo founder from India.
MCP servers are the connectors that let AI apps like Claude and Cursor use real products. I built QAmcp, a free open-source tester for them, and kept running into the same thing: a server can be up and still broken for an AI.
So before building monitoring, I re-tested 49 public MCP servers every hour for three days. None went down. But 6 changed their tools, 3 dropped in score, and on one server two tools vanished from the list and came back three times in two days, while it answered every check. An uptime check sees none of that.
The 15 checks in every test:
Connects the way an AI app does (the MCP handshake, not just “the link loads”)
Ready for the newest MCP version
Lists its tools
Every tool and input has a description the AI can understand
Tools are labeled safe or dangerous, and the label agrees with the name and description
No known hidden instructions in tool descriptions (quoted word for word if found)
Tools work with good inputs
Bad inputs get a clean error, not a crash or a hang
Tools work together (one tool’s answer feeds the next)
Fast enough, and answers small enough for an AI to read
No passwords or keys leaked in answers or errors
Follows MCP best practices
Resources and prompts work and are described
Login setup works (for servers behind a login)
Blocks requests from other websites (for servers on your own computer)
What Monitor adds on top:
“Down” and “back up” emails, plus slow or going up and down
Every tool change, line by line: tools added or removed, inputs changed, labels flipped
New sentences in tool descriptions, so a person reads them before the AI follows them
Score drops, answers changing shape, a login that broke, a certificate about to expire
Only tools labeled read-only are run, and every result is checked by code, never by an AI.
Free for 1 server (a full test once a day). Pro is $19 a month for the first 50 customers: 3 servers, hourly tests, an up check every 5 minutes, Slack.
I’d love to hear what breaks on your MCP servers most often.