QAmcp
GitHub
Coming soon · waitlist open

Every teammate gets only the tools they need

Guard sits between your team's AI apps and your MCP servers. Tick the tools each team may use. Everything else is blocked and hidden, secrets are hidden before the AI sees them, and every call is logged.

First 50 teams: $49 a month instead of $99, for up to 25 people. No card now. Guard is being built.

Guard · Rulessample data
TOOLSUPPORTENGINEERINGFINANCE
jira.read_ticketALLOWALLOWALLOW
orders.searchALLOWBLOCKALLOW
orders.refundBLOCKBLOCKALLOW
github.create_prBLOCKALLOWBLOCK
github.delete_repoBLOCKBLOCKBLOCK
What the AI just didtoday
10:42 · priya · jira.read_ticket · allowed
10:43 · priya · github.delete_repo · blocked
10:47 · sam · orders.search · allowed, 2 secrets hidden
The problem

Connect once, get everything

How it works

One address, one rulebook

  1. Point your AI apps at GuardOne address instead of one per server. Guard passes messages on to the real servers.
  2. People sign in with their work accountGoogle or Microsoft, so Guard knows who is asking.
  3. Tick the tools each person or team may useA simple page with switches. One switch makes a whole server read-only.
  4. Everything else is blocked and hiddenThe AI never sees a tool it isn't allowed. If it calls one anyway, Guard refuses with a plain message.
  5. Secrets are hiddenKeys, passwords and personal details are replaced with [hidden] before the answer reaches the AI.
  6. Every call is loggedWho used which tool, when, and what was blocked.
Sensitive data

Hidden before the AI sees it

Guard reads every tool answer on its way back and hides what shouldn't leave. This is a sample.

What the server sent

customer: priya@acme.com
api_key:  sk_live_9f2c41d8a7
card:     4242 4242 4242 4242

What the AI gets

customer: [email hidden]
api_key:  [secret hidden]
card:     [card hidden]
Price

A price you can see

Questions

Questions

Is Guard ready?

Not yet. It's being built. Join the waitlist and we email you once when it's ready to try.

Which servers does it cover?

The plan is every MCP server your team uses: ones you built, and outside ones such as GitHub, Jira, Slack and Notion.

Claude Code and Cursor already let me switch tools off. Why Guard?

Those switches are set by each person, on their own computer, one app at a time. Guard is one rulebook for the whole team and every AI app, with a log.

What can't Guard see?

Guard covers everything that goes through your MCP tools. A file an AI app reads straight from someone's computer doesn't pass through Guard.

How does it fit with QAmcp?

QAmcp tests a server and shows which tools are risky. Guard uses the same checks to suggest what to block.