QA.tech MCP server · for AI coding agents

Your coding agent can now
test what it builds.

Connect QA.tech's vision-based testing agents to Cursor, Claude Code, or any MCP client.
Your coding agent runs, creates, and validates tests without leaving the editor.

Get a demo

Trusted by high-performing engineering teams at:

Lovable
Nordea
Pricer
AstraZeneca
Upsales
Sambla Group
AdminPulse
MeetingPackage

What the MCP
server does.

Connect Claude Code, Cursor or Codex and your coding agent can list tests, start runs, read results and investigate failures without leaving the editor. Connections can be scoped to selected projects rather than your whole organisation.

You can also register your own remote MCP server, so the QA.tech chat agent can call your tools, with permissions set per tool.

Code shipped this weekat agent speed
Tests shipped this weekby hand
320h/mo
QA time saved

"We have replaced over 320h of manual testing every month with QA.tech."

Fredrik Seidl
Fredrik Seidl – CTO at Upsales
Upsales is a SaaS platform for CRM, marketing automation & Sales in B2B. ~150 employees.
Read their story

Things your agent now controls.

Sixteen tools, designed to live alongside the other MCP servers in your editor. Cursor's tool budget is finite – every one of these earns its place.

start_runlist_runsget_runcreate_chatcreate_test_caselist_application_environmentsget_environmentcreate_remote_tunnel+ 8 more

Run tests on demand

start_runlist_runsget_run

Your agent triggers a full test plan after it ships a change – or just the 3 tests touching the file it edited. Results stream back into the chat.

Run the checkout regression suite against the staging preview

Create tests from chat

create_chatcreate_test_case

Your agent describes a flow in plain English; QA.tech generates the test and adds it to your plan. Net new coverage with no test-author bottleneck.

Add a test for the new SSO login path I just merged

Manage environments

list_application_environmentsget_environment

Switch which environment your tests run against – staging, preview, prod – without touching the app. Useful for ad-hoc PR-style validation.

Run smoke tests against the preview URL Vercel just deployed

Tunnel to local dev

create_remote_tunnel

Open a secure tunnel so QA.tech agents can test your local dev server. Validate a change before you even commit.

Open a tunnel and run login against localhost:3000

What happens when your agent calls QA.tech.

  1. 01

    Your agent calls.

    Cursor, Claude Code, or any MCP client invokes one of QA.tech's 16 tools – start_run, create_chat, and others.

  2. 02

    QA.tech does the work.

    Tests run, get created, or get queried in our secure cloud. No infrastructure on your side. Our agents test what users see, not the DOM they happen to render.

  3. 03

    Results stream back.

    Pass/fail, screenshots, reasoning, links to full traces – delivered into the same chat your agent is in.

Works with

WebWeb
Mobile webMobile web
iOS AppsiOS Apps
Android AppsAndroid Apps

Use cases teams start with.

Once the agent is wired into your IDE or CI, these are the checks teams ask it to run first:

Testing AI-generated codePR & CI/CD gatesAI test generationShift-left testingSmoke testingAPI contract testingAutomated regression testingExploratory testing

Your team moves fast. Can your testing keep up?

QA.tech agents test your product autonomously, so moving fast never means shipping broken. See how it works in a 30-minute demo.

Get a demo

Frequently asked questions

Can't find what you're looking for? Reach out.

What is agentic testing, and how does MCP enable it?
Agentic testing is software testing run by AI agents that behave like real users – navigating the UI, exercising flows, validating outcomes – without anyone scripting selectors or maintaining brittle test code. The agent decides what to test, runs it, and reports back.

MCP (Model Context Protocol) is the standard that lets those agents be called by other agents – including the coding agent in your editor. With QA.tech's MCP server, Cursor or Claude Code can trigger runs, create tests, and pull failure analysis directly. The agents see your UI the way users do, so tests stay green through refactors instead of breaking on every selector change.
How is this different from other testing tools with MCP?
Two real differences. First, our agents are vision-based. They test what users see, so a UI refactor doesn't trigger a red build – the suite stays green because the experience stayed correct. Most other platforms still use selector-based automation underneath; the MCP layer is new, the testing engine isn't. Second, our agents are designed from the ground up to operate in an agent workflow. The same testing surface that runs on a cron also runs from your coding agent's chat, with the same project knowledge and the same context. We're not retrofitting AI access onto a scripted-automation product. See the head-on QA.tech vs Mabl comparison.
How do I get access?
MCP access is rolled out hands-on with our team. Book a demo and we'll scope your setup, get you onboarded, and configure the right clients for your stack – Cursor, Claude Code, or any other MCP-compatible editor. See the MCP server integration for what the connection exposes.
Does this need a separate plan or license?
No. If you have a QA.tech account, you have the MCP server. The same authentication, the same projects, the same runs.
What can the agent not do?
The agent can read your project, trigger runs, create tests, and manage environments. It cannot delete data, change billing, or modify users. We keep destructive actions behind the web app for now.
Is this secure?
Yes. Authentication is scoped per user; tool calls run with the same permissions as the user behind them. QA.tech is SOC 2 Type 2 compliant, and your data is never used to train models.

Latest changes: MCP – Connect AI coding tools to QA.tech with one command.