AI coding agents + Regresco
Regression testing for AI coding agents
Your agent writes code faster than anyone can click through the app before a release. Connect it to Regresco through MCP and it saves regression flows for your critical user paths, runs them in real Chromium in the cloud and reads what failed. The flows stay in your account and run again on a schedule and from CI.
Available on every plan, including Free.
Claude CodeTested
ChatGPTTested
CursorIn testing- CodexIn testing
Claude Code is a trademark of Anthropic. ChatGPT, Codex and the OpenAI logo are trademarks of OpenAI. Cursor is a trademark of Anysphere. Shown to name the clients Regresco works with; no endorsement implied.
- 1
Connect your AI assistant
regresco https://regresco.com/mcp
Done:Connected, Regresco tools available
- 2
Save a regression flow
create_flow "Checkout" 6 steps
Done:Saved in project Staging
- 3
Run in the cloud
run_tests flow: "Checkout"
In progress:Queued, then running on a cloud worker
- 4
Review the result
get_run_failures
Failed:Step 5 failed: order total not visible. Screenshot attached.
The gap
Why agent-written code needs a regression suite
An agent checks the change it was asked to make. The flows around that change need their own checks, run every time.
It checks what it changed
Ask for a new billing page and the agent tests the billing page. A shared component it edited on the way can break login or checkout without anyone opening them.
Browser checks end with the chat
When an agent drives a browser to look at a page, that check lives in the conversation. Nothing runs it again next week.
Unit tests miss broken pages
Unit and API tests can all pass while a button stops responding. A browser that walks through the real page catches it.
How it works
From one prompt to a suite that keeps running
Every step goes through Regresco's MCP tools, so the agent needs no browser setup of its own.
- 01
Connect the agent
Add https://regresco.com/mcp in Claude Code, Cursor, Codex or ChatGPT. Sign in with OAuth or use a personal token, and choose what the agent may do.
- 02
Save regression flows
The agent writes ordered browser and API steps from the real page or your description and saves them in a project that points at your staging URL.
- 03
Run in the cloud
run_tests queues the run. A cloud worker executes it in real Chromium with Playwright and keeps step results and screenshots.
- 04
Read what failed
The agent reads the failed step, the error, console errors and the screenshot. Regresco marks each failure as a regression, a broken locator or flaky.
Prompts
What to ask your agent
Start with the paths that would hurt most if they broke. Point every flow at staging, with test accounts and test payment details.
“Create a regression flow for the signup page on staging and run it in Regresco.”
create_project, create_flow, run_tests, get_run
“Run the checkout flow and tell me which step failed and why.”
list_flows, run_tests, get_run, get_run_failures
“Before I merge this branch, run every flow in the Staging project and summarize the failures.”
run_tests, get_run, get_run_failures
“Has the login flow been unstable in its recent runs?”
get_flow_stability
If the agent needs something Regresco does not have, such as a test account or a selector, it should ask you instead of guessing.
Compared
A browser session in chat vs a saved Regresco flow
Browser tools such as Playwright MCP let an agent drive a browser during a conversation, which helps when exploring a page or debugging. Regresco keeps the checks that matter as flows.
- Where the check lives
- In chat: In the conversation
- Regresco: In your Regresco project, editable in the dashboard
- Next release
- In chat: Ask the agent again, or turn it into test code in your repository and maintain it
- Regresco: The same flow runs from the dashboard, a daily or weekly schedule, or CI
- Browser
- In chat: Usually a browser on your machine
- Regresco: Real Chromium on Regresco's cloud workers
- History
- In chat: The chat log
- Regresco: Every run, compared with a baseline run
- When a step fails
- In chat: What the agent noticed during the session
- Regresco: Failed step, error, console errors, screenshot and a classification
Control
What the agent can and cannot do
The agent signs in as your Regresco account with the permissions you choose. Revoke a token or disconnect an app in Settings and its next request is refused.
With your permission, it can
- Read projects, flows, runs and results
- Create and edit projects (projects:write)
- Create and edit flows (flows:write)
- Start runs (runs:execute), one run of your monthly quota each
It cannot
- Delete projects or flows
- Change your plan or billing
- See another account's projects
- Read saved passwords and secret headers, which come back as [redacted]
- Open localhost or private network addresses
Can an AI coding agent write regression tests?
Yes. Through Regresco's MCP server, an agent such as Claude Code saves regression flows as ordered browser and API steps in your Regresco project. You can open and edit every flow in the dashboard, and the agent can update it later.
Which AI agents work with Regresco?
Any MCP client that connects to remote servers over Streamable HTTP. Claude Code and ChatGPT have been tested against the production server. Cursor and Codex have documented setups that have not been tested end to end yet.
How is this different from Playwright MCP?
Playwright MCP lets an agent control a browser during a conversation. Regresco saves the flows the agent creates, runs them on managed cloud Chromium, keeps every run and compares it with a baseline. You can use both: a browser tool to explore a page, Regresco for the checks that should run before every release.
Do I need to write Playwright code?
No. A flow is a list of steps such as navigate, click, fill and assert, stored in Regresco. The agent writes the steps and you can change them in the dashboard. If you already have Playwright tests, you can import them.
What does it cost?
MCP is included on every plan, including Free. Each run the agent starts uses one run of your monthly quota: 20 on Free, 200 on Pro and 1,000 on Business. Reading projects, runs and results uses none.
Is it safe to give an agent access to Regresco?
The agent gets only the permissions you approve, and nothing can be deleted through MCP. Saved passwords and secret headers come back as [redacted]. Screenshots show what the test browser saw, so use test data on staging. You can revoke access in Settings at any time.
Give your agent a regression suite
Connect Claude Code with one command, or ChatGPT with an OAuth sign-in. The free plan includes 1 project and 20 runs a month.