Skip to main content

Your AI agent now tests your application autonomously.

Squish MCP turns Cursor, Copilot, Windsurf, and Claude Code into autonomous testing partners. Agents explore and control your application live through the same Squish engine trusted by automotive, medical, and aerospace teams for over 20 years.

cursor · squish-mcp
> Generate a test that opens My Application and creates a new project.
◌ Querying live UI state via Squish MCP…
▸ launch_application("myapplication")
▸ find_focused_window() ▸ "Welcome"
▸ click("New Project")
◌ Dialog opened. Fetching active controls only.
▸ select_template("Sample Application")
▸ verify() ▸ ✓ Test passes

 
60%
Faster test case
generation
30%
Lower token
consumption
1000+
UI elements handled
per session
20+
Years of Squish
testing heritage
Our AI Strategy

Engineers already have AI tools they love. We meet them there.

Qt Group's Software Quality Tools AI strategy is built on three commitments. Squish MCP is the most direct expression of the third, and the reason your team can adopt AI-assisted GUI testing without changing the editor they already use every day.

01

Accelerated test development

AI working inside Squish to generate, debug, and improve test scripts faster, with full context.

02

Cross-product workflows

AI working between Squish, Coco, and TestCenter for value you can't get from any single product alone.

You are here
03

Embrace any environment

AI wherever your engineers already work: Cursor, Copilot, Windsurf, Claude Code. Model-agnostic, on-prem capable, no lock-in. This is what Squish MCP delivers.

What's New

From assistant to autonomous testing partner.

Squish MCP Version 0.2 is a major upgrade. AI agents can now explore and control an application completely on their own, without ever looking at the source code. The MCP server provides live application state and UI interaction tools, so the agent works the way a human tester would.

 

Autonomous exploration

Agents discover an application's structure by interacting with it live. They understand what your software does the way a human would.

 

60% faster, 30% leaner

Compact, on-demand state representation replaces bulky object snapshots. Agents fetch only what they need, when they need it, keeping context windows small and iterations fast.

 

Scales to complex applications

Real-world targets carry well over 1,000 UI elements. Version 0.2 handles that scale where Version 0.1 struggled, opening up MCP-driven testing to far more projects than before.

 

Goes beyond test generation

Squish MCP doesn't just write tests. Agents can triage failures, audit suite coverage against requirements, find duplicate test cases, and explore for hidden defects. Entire work packages, handled.

How It Works

Live application context, fed to your agent.

Squish MCP is a Model Context Protocol server. It plugs your AI code editor into a running Squish session, giving the agent both the live state of your application and the tools to act on it.

01 / PROMPT

You write the intent

"Generate a test that creates a new project and verifies the file is saved."

02 / AGENT

Agent calls MCP tools

Cursor, Copilot, Windsurf, or Claude Code requests UI state and actions through the Squish MCP server.

03 / SQUISH

Squish drives the app

The Squish engine inspects, clicks, types, and verifies inside the live application, exactly as it would in a normal test run.

04 / RESULT

Compact state returns

Only relevant context flows back to the agent, keeping iterations fast and tokens lean. The agent decides the next move.

Real-World Use Cases

Whole work packages, handed to your agent.

Squish MCP isn't just an autocomplete for test scripts. It can take ownership of full pieces of work, from creation to triage to suite hygiene, letting your engineers focus on the high-value testing decisions.

PROMPT
"Generate a test that opens My Application and walks through the New Project wizard to create a draft project."
Test Generation

Build complete tests, end-to-end

The agent launches the application, navigates the UI, writes the Squish script, runs it, and iterates until green. v0.2 handles 1,000+ UI element apps reliably, 60% faster than v0.1.

PROMPT
"Try to crash the application."
 
Exploratory Testing

Find defects no script would catch

Free the agent to probe edge cases, malformed inputs, and unusual interaction sequences. Discover stability issues your structured test suite was never written to look for.

PROMPT
"Is there a UI workflow with a dead end, where the user cannot get out of again?"
Quality Audit

Surface invisible risk surfaces

Orphaned dialogs, disabled-but-reachable controls, dead UI states: the kind of issues that ship to production unnoticed. The agent maps them out by exploring the UI graph for you.

PROMPT
"The test suites under /payments cover which functional requirements?"
Coverage Mapping

Map tests to requirements automatically

Critical for ISO 26262, IEC 62304, and FDA-regulated work. The agent reads your suite, queries the live application, and produces a traceable mapping of tests to functional requirements.

PROMPT
"Test case checkout_flow.py produces an error. Is the application buggy or is the test wrong?"
Failure Triage

Stop the manual triage loop

The agent reproduces the failure, inspects the live app state, compares it to the test's expectation, and tells you which side the bug is really on. Saves hours per failed run.

PROMPT
"Which test cases in /tests/ui are duplicates of one another, and why?"
Suite Hygiene

Clean up suites that grew organically

Long-lived projects accumulate near-duplicate tests as teams change. The agent reads, compares, and groups them, turning months of manual cleanup into a single afternoon's review.

PROMPT
"Can I cancel the wizard in each step?"
UI State Verification

Verify any state, on demand

Ad-hoc queries against the live UI without writing a script first. Useful when reviewing a feature, debugging a customer report, or sanity-checking a release candidate.

Compatible AI Code Editors

Already in your team's stack? Already supported.

Squish MCP follows the Model Context Protocol standard. If your editor speaks MCP, it speaks Squish.

Reference Setup

GitHub Copilot

Our fully documented, step-by-step integration. The Customer Portal walkthrough covers configuration, local-model setup, and your first prompt.

Also works with
Cursor
MCP-compatible
Windsurf
MCP-compatible
Claude Code
MCP-compatible

The GitHub Copilot walkthrough applies to any MCP-compatible client. Same pattern, same setup steps.

Security & Compliance

Built for regulated, air-gapped, and safety-critical environments.

We work with teams in automotive, medical, and aerospace where third-party LLM use is a non-starter. Squish MCP is engineered to give you the choice, including running entirely on your infrastructure.

Local-model capable

Run the MCP server with locally hosted LLMs (for example, GitHub Copilot's local model option), so no test data, source code, or application state ever leaves your network.

Model-agnostic

Choose the model that fits your compliance, cost, and capability requirements. No lock-in to a single provider, and no compromise on the AI strategy your organization has already settled.

Data stays with you

When configured for local deployment, your application state, test scripts, and execution data remain inside your infrastructure. Nothing flows to external services unless you choose it to.

Stop writing tests one keystroke at a time.

Bring the AI tools your team already uses into the Squish testing engine your organization already trusts.