Your AI agent writes your regression tests. Delta‑QA proves them.
Connect Claude Code, Cursor or any MCP client to Delta-QA. Your agent creates the scenario and writes its steps; Delta-QA replays them in its own browser, on your real site, photographs every step and only validates what actually worked.
›Create a regression test of the “Checkout” journey on example.com
list_workspaces…1 workspace, “Can edit” rolecreate_scenario…Scenario created, step 1 opens the pageDRAFTset_scenario_steps…3 steps saved, #2 to #4capture_scenario…Replay started on the real siteget_capture_resultReplaying, taking photos…Action #4:button.paymatches no visible elementFAILED- The Pay button has no such class: I'll target its data‑testid attribute instead.
set_scenario_steps…Step #4 fixed:button[data-testid=pay]capture_scenario…New replay startedget_capture_resultReplaying, taking photos…4 steps replayed, 4 photosCOMPLETEDactivate_scenario…The capture becomes the referenceACTIVE
- #1
- #2
- #3
- #4
What your agent does with Delta-QA
The Model Context Protocol (MCP) is the open standard through which an AI agent discovers tools and calls them. Delta-QA's MCP server exposes eight: create a scenario, write its steps, replay it, read the verdict, fix, validate. It is the path of the scenario screen, without the screen.
You ask in plain words: “Create a regression test of the sign-up journey on https://…”. The agent writes the steps, Delta-QA replays them on the real site and photographs each one. Once validated, the scenario is replayed on every release, like a scenario recorded by browsing.
Connect your agent in three steps
- Create your free account and your workspace.
- Create an API key in Settings, Workspace tab, API keys section: a name, the “Can edit” role for an agent that writes scenarios, an expiry. The key is shown only once, with the ready-made command.
- Paste the command in a terminal for Claude Code, or give your MCP client the server address and the header below.
claude mcp add --transport http delta-qa https://api.delta-qa.com/api/mcp --header "Authorization: Bearer dqa_…"- Server address
https://api.delta-qa.com/api/mcp- Authentication header
Authorization: Bearer dqa_…
The server speaks the protocol's “Streamable HTTP” transport, stateless: every call is a request of its own. Any MCP client that accepts a remote server and an authentication header can connect, from Cursor or VS Code to your own agent.
The eight tools of the server
They follow the order of the journey. An agent discovers them on its own: there is nothing to explain to it.
| Tool | What it does | Role required |
|---|---|---|
list_workspaces | Lists your workspaces, with your role in each. | Read-only |
list_scenarios | Lists the scenarios of a workspace, most recent first. | Read-only |
create_scenario | Creates a draft scenario from a name and a start URL; step 1, opening the page, is set by Delta-QA. | Can edit |
get_scenario | Reads a scenario and its steps, numbered as the replay numbers them. | Read-only |
set_scenario_steps | Writes every step of the journey, out of fourteen gestures: click, type, select, key, scroll, navigate… | Can edit |
capture_scenario | Replays the scenario in Delta-QA's browser, on the real site, and takes its photos. | Can edit |
get_capture_result | Reads the verdict of the replay; a failure names the step to fix. | Read-only |
activate_scenario | Validates the scenario: the capture becomes its reference, compared on every release. | Can edit |
A step, as the agent writes it
Every step is a gesture and its target: a CSS selector that matches exactly one visible element at that point of the journey.
{"action": "CLICK", "selector": "[data-testid=add-to-cart]", "text": "Add to cart"}
{"action": "FILL", "selector": "#password", "value": "…", "sensitive": true}
{"action": "KEY", "value": "Enter"}
{"action": "SCROLL", "value": "800", "photo": false}- The visible text of the element helps Delta-QA find it again when the site changes.
- A sensitive value (password, card) is encrypted as soon as it arrives, typed at replay, veiled on the photos and never read back.
- One photo per step by default: that is what every future check compares.
- Nothing is validated on trust: a step that does not replay fails the capture, which names the action at fault.
One key per workspace, under your control
- It belongs to the workspace, not to a person: it only acts in that workspace and outlives the departure of whoever created it.
- Two roles: “Read-only” reads the scenarios and their checks without starting anything; “Can edit” creates, replays and validates.
- A mandatory expiry: 30 days, 90 days or 1 year, never longer. Revocation is immediate, from the same screen.
- Shown only once, kept as a SHA-256 fingerprint, never logged. A revoked or expired key is refused.
- The same doors as the interface: same roles, same workspaces, same quotas as a person. Available on every plan, free account included.
The agent proposes, Delta-QA proves
An AI agent can pick the wrong selector or claim that a journey works. With Delta-QA, nothing is validated on trust: every step is replayed in a real browser, on the real site, and photographed. A failing replay names the step to fix; only a passing replay becomes the reference.
From then on, every release is compared with that reference by Delta-QA's deterministic AI, not an LLM: the same difference always gets the same verdict, and every difference can be explained.
Automated regression testing, explained · What Delta-QA detects