WebMCP forms can now pass: reading the Lighthouse 13.5 audit changes

A form supports WebMCP, yet its audit result says “not applicable.” Lighthouse 13.5 includes a change that can turn that result into a pass.[1][2]

Lighthouse is a tool for auditing websites. This update makes a site's form coverage for WebMCP easier to interpret in the audit output.

But a passing audit does not guarantee a successful AI-assisted purchase. Here is what changed and what teams should check separately.

What changed in WebMCP on September 17 and 18?

The WebMCP specification document and the Lighthouse audit update are separate developments. Distinguish the document dated September 17, 2026, from Lighthouse 13.5, released September 18.[1][2]

Date	Subject	Confirmed change or status
September 17	WebMCP document	Published as a Draft Community Group Report
September 18	Lighthouse 13.5	Form-coverage result change and a warning about registered tool counts

WebMCP lets websites expose available operations as tools for AI. The document is a draft, not an announcement that WebMCP has become a W3C Standard.[1]

These are recent changes checked on September 26, but they did not occur within the preceding 24–48 hours. The release date and the date we cover it are different.

For the broader context, start with our WebMCP guide.

When does WebMCP form coverage change from n/a to pass?

When all relevant forms are annotated for WebMCP, the change reports a pass instead of n/a. Here, n/a means not applicable.[2][3]

The annotations describe a form's operation for AI. The change makes complete coverage visible as a passing audit result.

Partially uncovered forms do not automatically become a failure

The Pull Request explains that this change does not make forms with missing coverage fail across the board. “Pass when all are covered” and “fail when any are uncovered” are different rules.[3]

State	What this change establishes
All relevant forms covered	The result becomes pass rather than n/a
Some forms uncovered	This change alone does not establish an automatic fail

Comparing an older n/a result with a newer pass does not prove that the implementation improved. The reported result can change with the audit version even if the website is unchanged.

Record the Lighthouse version as well as the date of any site changes when making comparisons.

Too many registered WebMCP tools can trigger a warning

Lighthouse 13.5 also adds a warning when the number of registered WebMCP tools exceeds a recommended limit. This is separate from the form-coverage change.[2][4]

It does not mean the WebMCP specification imposes a universal hard limit. An audit recommendation and a specification constraint are different things.

If you see the warning, inspect your registered operations. These are our suggested review questions:

• Is the same operation registered more than once under different names?
• Are tools registered for operations unavailable on the current page?
• Do the names and descriptions make the differences clear?

For example, product search and stock lookup have distinct purposes. Two operations that run the same search under the same conditions may be candidates for consolidation. This is an illustrative example, not a claim that Lighthouse verifies the correctness of their execution.

Check both the audit version and the actual operation

Record the Lighthouse version, interpret the audit, and then test the operation itself.

At release, Lighthouse 13.5 was announced for Chrome 156 DevTools, with PageSpeed Insights integration still forthcoming. Reading the release notes does not establish that every environment you use already runs 13.5.[2]

Check the version shown in the report first. Then record the URL, audit date, form-coverage result and any registered-tool warning.

A pass does not verify agent identity or purchase success

A passing WebMCP form audit does not tell you who the agent is or who approved an order. Identity verification, such as Web Bot Auth, and the site's operation and approval design require their own checks.[5]

For functional testing, use defined test conditions to check whether product search works and returns appropriate results. If orders are involved, use a test environment to inspect quantities, totals and the point at which approval is requested.

Keep the diagnostic result and the functional test record together. That makes it easier to explain exactly which parts of your WebMCP implementation have been checked.