The AI agent and MCP

The agent

Open Agent in a project's sidebar, or press ⌘J anywhere to open it in a side panel. Ask in plain English, for example:

• "Which requirements have no test cases?"
• "Triage the failures in the latest CI run."
• "Create a regression run with all critical checkout cases and assign it to me."
• "File a high-severity bug: coupon not applied on mobile Safari."
• "Is release v2.4 ready to ship?"

The agent reads your project data and can take actions. Actions follow the same rules and permissions as the app: for example, it can't change results on a completed run, it can't complete a run until every case has a result, and viewers can't make changes through it.

Approvals

Some actions change or remove things that already exist, or send something outside sqale:

• updating or linking a bug;
• recording a result, or completing or assigning a run;
• editing or moving test cases;
• renaming or deleting a suite;
• posting to Slack or commenting on Jira.

The agent can only propose these. You see a card describing the exact change (for example Update bug "Coupon not applied": status open → fixed) with Approve and Cancel. Nothing happens until you approve, and it runs on exactly the record shown. A proposal expires after 30 minutes. Creating new things (bugs, runs, requirements, suites, test cases) happens straight away.

Activity

The Activity tab lists everything the agent changed in the project: from chat and from automations, by whom and when, and every proposal with its outcome. It also shows what the agent remembers about you: preferences it learned from your own chats in this workspace. Only you can see it, and you can clear it at any time.

Add product context under Project settings → General: describe what your app does, who uses it, and its key flows. Answers get much better with it.

Automations

The Automations tab runs the agent on a schedule ("Every weekday at 9:00, summarise yesterday's CI failures") or when a webhook is called.

• Who it runs as: an automation runs as the person who set it up, or who last changed what it does.
• What it can do: it can read everything, and create bugs, runs and test cases. It can't change or delete existing records, post to Slack or comment on Jira, because nobody is there to approve; it reports what should change instead.
• Slack: turn on Post each run's summary to Slack to get the result there.
• History: each schedule and webhook shows its recent runs, with the summary and the changes each made.
• Worker: automations need sqale's background worker to be running.

Use sqale from Claude Desktop, Cursor and other MCP clients

sqale is an MCP server, so AI tools that support the Model Context Protocol can use your sqale data directly.

1. Settings → AI assistant integration (MCP): create a key (Pro plan).
2. Add sqale to your client. The settings page shows ready-to-paste configuration:
   • Cursor and other clients that support remote servers: the server URL https://sqalehq.ai/api/mcp with the header Authorization: Bearer <your MCP key>.
   • Claude Desktop: its config file starts local servers only, so the configuration runs the small mcp-remote bridge with npx (needs Node.js installed).
3. Ask your client things like "list the failing test cases in project SHOP".

Every MCP call is recorded in an audit log.

Privacy

The agent is powered by Anthropic Claude. Your data isn't used to train AI models.