1. Docs
  2. AI
  3. MCP & AI assistants

MCP & AI assistants

The SmartestQA MCP server lets an AI assistant such as Claude work in your SmartestQA projects for you: find out where a project stands, plan and write test cases, run them, read results and handle defects. It works with your own account and permissions, and it changes nothing until you approve the change.

6 min read Updated CustomersAdmins

In short

On this page you can

  • Connect Claude on claude.ai, Claude Desktop, Claude Code, Cursor or another MCP client.
  • Ask in plain words where a project stands and what to do next.
  • Let the assistant plan, write and dry-run test cases with you.
  • Read red runs, heal locators, handle defects and Jira from the chat.
  • Approve every change as a diff before anything is written.

How to use it

  1. Pick a way to connect: the claude.ai connector, the Claude Desktop extension, or the one-line install.
  2. Connect with your SmartestQA sign-in or your host’s API key (never a password in the chat).
  3. Ask “Which SmartestQA projects can I access?” and choose a project.
  4. Tell the assistant what you want; it proposes a change and shows the diff.
  5. Say yes to apply it, or ask for a change to the draft.

At a glance

What
An MCP server: a standard way for an AI assistant to call SmartestQA tools. 18 tools, 4 ready-made prompts.
Who
Every role. The assistant can do only what your user can do. Removing records needs an Admin and a click.
Connect
claude.ai: Settings › Connectors. Claude Desktop, Claude Code, Cursor: Device FarmTest AgentsMCP.
Sign-in
Connector: your SmartestQA sign-in, never shared with Claude. Others: the API key of your host in Test Agents.
Changes
Two steps: smartest_propose shows a diff and writes nothing; smartest_apply writes it after you approve.
Languages
The assistant answers in your language and writes into SmartestQA in the project’s language.
Project types
WEB, MOBILE and DESKTOP. Step rules exist for web, mobile and desktop steps, plus API, database, mail and load tests.

Start from your question#

What MCP is#

MCP (Model Context Protocol) is an open standard that lets an AI assistant use tools of another system. The SmartestQA MCP server offers tools such as “where does this project stand”, “propose a change”, “write and dry-run this scenario” or “why did this run go red”. The assistant decides which tool to call; SmartestQA checks every call against your permissions and its own rules, and answers with what happened and what usually comes next.

Ways to connect#

WayWorks inSigns in withInstall
Connector (remote)claude.ai, and Claude apps signed in to the same Claude accountYour SmartestQA email and password, on SmartestQA’s own pageNothing: add a URL in Claude
Claude Desktop extension (.mcpb)Claude Desktop on Windows or macOSYour host’s API keyDownload, drag into Claude, paste the key
One-line installClaude Code, Cursor, Claude Desktop and other MCP clients; Windows, macOS, LinuxYour host’s API key (already in the command)Run one command; needs Node.js

Where it fits#

An assistant can help at every stage of the order of work, from describing the app to reading reports and healing locators.

  1. 1DiscoverPages, actions, notes
  2. 2AnalyzeModules, requirements
  3. 3AutomateScenarios, dry runs
  4. 4ScheduleModule pipelines
  5. 5ReportRuns, logs, defects
  6. 6MaintenanceHeal decisions

Buttons in the app hand work to the assistant: With MCP on the Setup Roadmap and Copy MCP prompt in Test Health copy a ready request you paste into the chat. See Commands from the app.

The order the assistant works in#

The server tells the assistant to build a project in this order, and to move on only when the step before is in place:

  1. The environment (project language, start URL, where runs happen) and a source of knowledge (code, browser, documents or what SmartestQA already knows).
  2. Modules, then each module’s main requirements (MAINs), then their sub-requirements (SUBs).
  3. A scenario plan for each SUB, then each scenario written and dry-run until it is green. After three red runs a scenario is BLOCKED and the assistant brings you the options instead of trying again.
  4. The module’s API inventory, API and security scenarios, then its pipelines.
  5. Load tests of critical endpoints, then the status check until its “done” list is complete.

The building blocks#

MCP client
The app the assistant runs in: claude.ai, Claude Desktop, Claude Code, Cursor or another client that speaks MCP.
Connector
The hosted SmartestQA MCP server that claude.ai connects to over the internet, with a sign-in. Add it.
API key
The key of your host in Device Farm › Test Agents. The local MCP uses it to act as you. More.
Tool
One thing the assistant can ask SmartestQA to do, such as smartest_status. All 18.
Proposal and preview
A change shown as a diff, with an id (the preview). Nothing is written until the preview is applied. More.
Work key
A short name for a kind of open work, such as SCENARIOS (sub-requirements without a scenario) or HEAL (red pipelines). List.
Prompt
A ready-made request in the client’s menu: Status, Modules, Scenarios, Red runs. More.

Before you start#

WayYou need
ConnectorA SmartestQA account with a password (you can sign in with email and password); a Claude plan that lets you add custom connectors.
Claude Desktop extensionClaude Desktop installed; access to Device Farm › Test Agents to copy your key.
One-line installNode.js 18 or newer; Claude Code, Cursor or Claude Desktop opened once on that machine.
AnyAt least one project you can open. Test logins stored as data source keys in SmartestQA, not in the chat.

Best practices#

Do

  • Name the project in your request: “In project Web Shop, …”.
  • Read the diff before you say yes. Ask “what does this change touch?” if unsure.
  • Work module by module: one module, its requirements, its scenarios, its pipelines.
  • Give the assistant a source: your code folder, the running app, or a document.
  • Fill passwords and tokens yourself in SmartestQA, in the data source keys it creates.

Don’t

  • Paste a password, API key or token into the chat.
  • Approve “remove” changes in bulk without reading the list.
  • Ask it to keep retrying a BLOCKED scenario: decide with it what to change.
  • Share your API key with a colleague: each user has a host and a key of their own.

In this section#

Didn’t find your answer?Write to support with your project name and what you see on screen.

info@smartestqa.ai
© 2026 Selenity Technology · SmartestQAsmartestqa.aiThis page as MarkdownChecked against smartest-mcp-server 1.31.1 (b95acc0, src/next, core/registry), smarttest-fe predev 2a8c79cc (ConnectToClaude, OAuthConsent, downloadAssets, .env.production), smarttest-be 633c2a03 (oauth, ApiKeyAuthenticationFilter)