How QAmcp sees a broken serverOur own test server, broken on purpose · 38/100 · FAIL

SoonMonitor: hourly tests, an email when your server breaks

Test any MCP server, right here.

Name: QAmcp
Author: Abhiram Reddy

QAmcp connects to your server the way an AI app does, checks every tool, and draws you exactly where it breaks.

Paste your server link · nothing to install

No link handy? Try a real server: Context7 · DeepWiki

Free and open source
Never runs tools that could change data
Sends nothing anywhere but your server; logins stay on your computer

Found on this serverFull report →

1fetch_notes hides an instruction telling the AI to read your SSH key

2slow_report never answers, even to bad input

3slow_report has no description, so the AI has to guess

4delete_account isn't marked dangerous. Dashed boxes weren't run, on purpose.

5search_users shows a database password in an error

FOR TEAMSPublish a server for customers? Check it on every change

SHEET1 of 6 · v0.1.14 · MIT

Every check is a plain fact you can verify.

No AI judges anything. Every result is checked by code and shows its evidence (pass or fail, a time in milliseconds, an exact match, or the word that set it off), and the report always says what it could not test.

Check	What it means	On the test server
No hidden instructions	Tool descriptions don't smuggle orders to the AI, invisible characters, or tricks.	Failed: fetch_notes
No secrets leaked	Error messages don't show passwords, keys, tokens or internal file paths.	Failed: search_users
Tools work with good inputs	Each safe tool answers in time with something real.	Failed: slow_report
Bad inputs give a clean error	Missing or wrong-type inputs get a clear refusal, not a hang or a crash.	Failed: slow_report
Every tool is described	Descriptions and inputs explain themselves, and no two tools read the same.	Failed: slow_report
Tools are labeled safe or dangerous	Read-only and destructive labels exist, and match what the tool's name says.	Note: delete_account
Fast enough	Connecting, listing and answering, timed.	Note: search_users, 3.0 s
Login setup follows the spec	An AI app can find your login and sign in the way the MCP rules describe.	No login on this server
Resources and prompts work	What the server offers besides tools opens cleanly.	None offered
Tools work together	An ID one tool returns is accepted by the next tool, the way AI apps chain them.	No chains on this server
Ready for the newest MCP version	Protocol version and transport, checked against the current spec.	Note: older version

Check

What it means

On the test server

No hidden instructions

Tool descriptions don't smuggle orders to the AI, invisible characters, or tricks.

Failed: fetch_notes

No secrets leaked

Error messages don't show passwords, keys, tokens or internal file paths.

Failed: search_users

Tools work with good inputs

Each safe tool answers in time with something real.

Failed: slow_report

Bad inputs give a clean error

Missing or wrong-type inputs get a clear refusal, not a hang or a crash.

Failed: slow_report

Every tool is described

Descriptions and inputs explain themselves, and no two tools read the same.

Failed: slow_report

Tools are labeled safe or dangerous

Read-only and destructive labels exist, and match what the tool's name says.

Note: delete_account

Fast enough

Connecting, listing and answering, timed.

Note: search_users, 3.0 s

An AI app can find your login and sign in the way the MCP rules describe.

No login on this server

Resources and prompts work

What the server offers besides tools opens cleanly.

None offered

Tools work together

An ID one tool returns is accepted by the next tool, the way AI apps chain them.

No chains on this server

Ready for the newest MCP version

Protocol version and transport, checked against the current spec.

Note: older version

A report page you can send to anyone.

Add --html report.html and QAmcp writes one file: the score, every failure and note, everything it did not run and why. It opens offline, on any computer.

The server we broke on purpose38 / 100 · FAIL
Context7, a public server by Upstash100 / 100 · PASS

Context7 was tested independently on 15 Oct 2026, the same way anyone can. QAmcp is not affiliated with Upstash or Context7.

Server behind a login? Add --login. Its own login page opens in your browser, and the login stays on your computer.

A program on your computer, not a link? Use --stdio "npx -y your-server".

The report page for the broken server: score 38 out of 100, FAIL, with six failures and ten advisories listed.

Publish a server for customers? Check it on every change.

When your MCP server breaks, your customers' AI agents break with it. QAmcp runs in GitHub on every pull request, comments with the score, and fails the build if the server got worse or a tool disappeared.

Never runs risky toolsOnly tools labeled read-only, whose names and descriptions agree.
Sends nothing elsewhereNo tracking, no account, no AI. Only your server hears from it.
Your login stays yoursYou sign in on your server's own page; nothing is stored anywhere else.
Open sourceMIT licensed. Read every check before you trust it.

# .github/workflows/qamcp.yml
on: pull_request
jobs:
  qamcp:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
      - uses: QAmcp/qamcp-cli@v0
        with:
          url: https://your-server.com/mcp
          fail-under: 80

Try it on your server now.

Your server changed at 2:14 am. You'd know.

QAmcp Monitor tests your server every hour and sends you an alert like this the moment something breaks: it goes down, the score drops, a tool changes, or a secret leaks.

Founding price: $19/month for life for the first 50 teams (normally $29). We'll email you once when it opens. Nothing else, never shared.

Questions about QAmcp

What is QAmcp?

A free, open-source tester for MCP servers. It connects to a server the way an AI app does and tells you, in plain words, what is broken: connection, tool descriptions, safety labels, hidden instructions, bad inputs, leaked secrets and speed. Every result is checked by code, never by an AI, and shows its evidence.

Is it free?

Yes. The browser test, the command-line tool (npx qamcp-cli), the GitHub check and the directory are free. The code is open source under the MIT license.

Does it send my data anywhere?

No. It talks only to the server you test, and to its login page if it has one. There is no tracking and no telemetry. On the website, the test runs on our test service and the report is kept so you can share its link.

Can it change or delete anything on my server?

By default it only runs tools the server labels read-only, and it never runs tools that sound destructive, like delete_account, in any mode. With --look-only it calls no tools at all.

Can it test local servers, or servers behind a login?

Yes. Local servers from the terminal: npx qamcp-cli --stdio "your start command". Servers behind a login: add --login to sign in through your browser, or --header with an API key.

How is the score calculated?

Each check is worth points: a pass earns all of them, a warning half, a failure none. Checks that could not run don't count either way, and serious problems such as hidden instructions or leaked secrets cap the score. The full rules are on the How it works page.

How is it different from MCP Inspector?

MCP Inspector, the official tool, lets you connect and call tools by hand, which is ideal for exploring and debugging. QAmcp runs a fixed set of checks automatically and gives a score you can repeat on every change. They work well together.

My server is in the directory. Can I have it removed?

Yes. Open an issue on GitHub and we'll take it off the list. Scores in the directory are never changed for money.

Who makes QAmcp?

Abhiram Reddy, a solo founder, building it in the open. Found a wrong result? Report it on GitHub: being fair to every server matters most.