Cypress Cloud MCP: What’s New in September 2026

On September 15, 2026, Cypress announced a Cypress Cloud MCP update that adds per-spec and per-test retrieval, status filters, durations at four levels, commit and CI metadata, and deeper links into Accessibility and UI Coverage reports. Responses are compact by default. Cypress says existing connected users do not need to change their setup: the update is live for everyone already connected. Cypress’s announcement and its release index describe the update.

What changed

Cypress Cloud MCP connects an AI coding assistant to Cypress Cloud run data. The September update helps an assistant move from a run to its specs and individual test results while requesting compact responses by default. The announced changes are:

Change	What it provides	Useful for
Compact responses by default	Tools return a smaller initial response; an agent can request additional fields when needed.	Keeping an investigation focused on relevant run details. Cypress publishes no token-savings or performance benchmark for this change.
`cypress_get_specs`	Per-spec details for a run.	Finding which specs ran, their statuses, durations, or skipped and pending counts.
`cypress_get_tests`	Test results within a run.	Inspecting individual passed, failed, flaky, skipped, or pending tests, as available in the run.
Status filters	Filters for run, spec, or test status.	Narrowing a query to failures, skipped tests, or another requested status.
More durations and run context	Duration at run, spec, test, and attempt level; runs also include `commit_author` and `ci_build_id`.	Tracing a slow run to a spec, test, or attempt and relating it to a commit or CI build.
Deeper Accessibility and UI Coverage links	Links into report views and rule failures.	Following a finding into its supporting report during triage.

These are capabilities Cypress describes. The announcement does not report independently measured productivity, speed, or token improvements.

How to use the new details

The practical workflow is to identify the relevant run, ask for the level of detail needed, then narrow the result set. For example, a release report can connect a passing run on main to its commit author, specs, and skipped tests. A performance review can move from total run duration to the slowest spec and then to tests inside that spec. For failure analysis, ask for each attempt’s status, duration, and error so the assistant can distinguish a repeatable failure from one that passes on a later retry.

Prompts to try

These examples assume Cypress Cloud MCP is connected and the project is configured. Cypress published the first prompt as an example in its announcement; the other workflows are based on the new levels of data and filters described there.

For the latest passing run on main, give me a release report: who authored the commit, which specs ran, and which tests were skipped.

For the latest run on my branch, list failed and flaky tests. Include each test's status, duration, and the status, duration, and error for every attempt.

For the latest Cypress run, use spec status and duration to identify the slowest specs. Break the slowest one down by test and show the durations.

For the latest run on my branch, list skipped and pending tests grouped by spec.

To investigate an accessibility issue or a UI Coverage gap, ask for the relevant report and rule or view, then follow the returned deep link to the supporting report. Cypress’s documentation includes additional prompt guidance and examples in its Cloud MCP documentation.

Setup and availability

Cypress Cloud MCP became generally available on May 20, 2026. Cypress documentation says it is included on every Cypress Cloud plan at no additional cost. The September update is live for users who already connected; Cypress says no update is needed for them. Cypress’s general availability announcement and the setup documentation cover enabling and connecting it.

Have access to the Cypress Cloud organization and project whose run data the assistant should inspect.
Enable Cloud MCP from the organization’s Integrations page, as described in Cypress’s setup guidance.
Configure a compatible AI client to connect to https://mcp.cypress.io/mcp.
Use OAuth, Cypress’s recommended authentication method. Personal access tokens remain available for clients that require them.
Ask a narrowly scoped question that identifies a branch, commit, run, or run URL and the result level you need.

Exact client configuration fields depend on the AI client, so use that client’s current remote MCP and authentication instructions alongside Cypress’s setup page. Do not put a personal access token in a prompt or commit it to a repository.

What the update changes in day-to-day triage

Move from run to spec to test

A run-level summary can tell you whether a run passed or failed. The new spec and test tools let an assistant continue into per-spec details and individual test results. State the target run and ask for the needed detail, rather than requesting every result for every run at once.

Use status filters to answer a specific question

Ask for the status category that matters: for example, failed tests, skipped tests, or flaky tests. A request to list every test can be much broader than a request for all skipped and pending tests grouped by spec.

Use duration at the right level

Run duration gives an overall view. Spec duration narrows the search to part of the suite; test duration narrows it further; attempt duration helps compare retries. These levels help frame an investigation, but duration alone does not explain why a test is slow. Check the test, its attempts, and any available failure context before deciding on a fix.

Connect a run to its source and CI context

commit_author and ci_build_id add context to run results. They can help a release report identify who authored the commit and which CI build produced the run. Cypress’s release example asks for the commit author, specs that ran, and skipped tests for the latest passing run on main.

Follow report links for Accessibility and UI Coverage

The update adds deeper links to report views and rule failures surfaced by Accessibility and UI Coverage tools. Use the assistant to find and summarize relevant findings, then follow the report link when you need to inspect the supporting view or rule failure directly.

Edge cases and limits

The assistant needs the right connection and project access. Prompts cannot return data that the connected account or project does not make available.
“Latest” can be ambiguous. Specify a branch, commit, project, or run URL when several runs could qualify.
Large suites may require multiple calls. Cypress notes that large suites take several calls to page through; scope the request to a project or group when possible.
Compact does not mean complete. If a summary omits a field you need, ask the agent to retrieve the relevant spec, tests, or additional fields.
A run-level duration can hide a local bottleneck. Drill down to spec, test, and attempt duration before attributing the delay.
A report link is a route to evidence, not a diagnosis. Review the linked Accessibility or UI Coverage detail before acting on an agent’s summary.
This is Cloud run data. Cloud MCP provides Cypress Cloud context; it is not the same as attaching an assistant to a live local browser session.

Troubleshooting

Symptom	Likely cause	What to do
The assistant cannot find Cypress Cloud tools.	The MCP server is not connected in the client, or the organization has not enabled the integration.	Check the organization’s Integrations page and the client’s connection to `https://mcp.cypress.io/mcp`; follow Cypress’s setup documentation.
Authentication fails or the assistant cannot access a project.	The user has not completed the configured authentication flow, or lacks access to the organization or project.	Reconnect with OAuth where supported, or use a personal access token only where the client requires it. Confirm the account can access the target project in Cypress Cloud.
The answer describes the wrong run.	“Latest run” is ambiguous across branches, projects, or concurrent builds.	Specify the branch, project, commit, CI build, or paste the run URL.
The answer omits test or spec fields.	The first response is compact, or the prompt did not request the detail.	Ask for `cypress_get_specs` or `cypress_get_tests` details and name the fields, such as status, duration, or attempt error.
A suite-wide query is incomplete or slow to review.	Large suites may be paged through multiple calls, and a broad result can exceed what is useful for the question.	Filter by status, project, group, branch, or run, and ask for one level of detail at a time.
A duration points to a slow area but not its cause.	Timing identifies where time was spent, not the underlying reason.	Drill into the slow spec, test, and attempts; inspect the relevant test or replay context using the available Cypress workflow.
An Accessibility or UI Coverage summary lacks enough evidence.	The summary is not the full report detail.	Ask for the deep report link and the specific view or rule failures, then inspect the linked report.

Performance, reliability, and cost

Compact responses are intended to return details when needed instead of including every field on every call. Cypress publishes no numeric token-savings, latency, or performance benchmark for the September changes, so treat that as a response design choice rather than a measured guarantee. Large suites can require multiple calls; filtering to the target run, status, or project makes the investigation more focused.

The new metadata and per-attempt durations add context for reviewing runs, but they do not replace checking the underlying result or report. Availability of useful results depends on a connected and authorized client and the data accessible in the selected Cypress Cloud project. Cypress documentation states that Cloud MCP is available on all Cypress Cloud plans at no additional cost; plan pricing and other account terms are separate from this MCP update.

ScreenshotNeo for page captures

Cypress Cloud MCP helps an assistant inspect test run data. When the separate task is to capture a website as an image or PDF, ScreenshotNeo is the alternative to try first: it is a website screenshot API and MCP server, and only clean shots are billed. It complements run triage; it does not replace Cypress’s run, spec, or test data.

Or skip the browser setup

ScreenshotNeo takes a URL in one API request and returns a screenshot. See the ScreenshotNeo API documentation. This runnable cURL example saves a WebP capture of Stripe:

curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://stripe.com -o shot.webp

Cookie banners are accepted like a visitor and removed along with supported newsletter popups and chat widgets before the shot. Bot checks, blank pages, timeouts, failed loads, and cache hits cost nothing; response headers identify the page verdict and billing status. An MCP server gives Claude, Cursor, and other MCP clients screenshot tools. The free plan includes 1,000 screenshots a month with no card; paid plans start at $5 for 3,000.

FAQ

Do existing Cypress Cloud MCP users need to update anything?

No. Cypress says the September update is live for users already connected.

What do the two new tools retrieve?

cypress_get_specs retrieves per-spec details for a run; cypress_get_tests retrieves test results within a run.

Does the update publish a token-savings figure?

No. Cypress describes compact responses as the default but publishes no numeric savings benchmark for the September update.

Is Cypress Cloud MCP the same as running Cypress tests?

No. It connects an AI assistant to Cypress Cloud run data. The update concerns retrieval and investigation of recorded results.

Where can I find the setup steps?

Use the official Cypress Cloud MCP documentation for connection and client guidance.