Description of the bug

There are several points to note regarding the CruX data using performance_start_trace:

• It is not accurate: there are noticeable differences compared to the CruX data returned by the PageSpeed website. Using the PageSpeed Insights API and a Google Console API key, we can retrieve the exact data reflected in the CruX.
• The FCP value is not being returned
• Errors are not being reported when access to CrUX fails: for example, I ran a performance_start_trace on a website that does have CrUX data on the PageSpeed Insights website, but performance_start_trace returned: Metrics (field / real users): n/a – no data for this page in CrUX\ nAvailable insights:\n. When I tried a second time from another AI agent, it did return the data. There must have been a connection or memory error, etc., that wasn’t reported; it simply returns “no CrUX data”

Is there a solution for this? We need to be able to automate performance audits, but whoever performs them must also cross-check the data with PageSpeed Insights... if these data differ significantly, it’s difficult to know whether the results returned by the MCP + automated performance agent are accurate or not.

Thank you

Reproduction

1. Use Claude + MCP Devtools
2. Tell them to visit a website, run a performance_start_trace, and send you the CruX data.
3. Copy that URL and submit it to PageSpeed Insights
4. Compare the results from the CruX and you'll see that they're inaccurate
5. The FCP metric is not available

Expectation

We need the CruX data to be almost as accurate as the data returned by PageSpeed Insights so that we can ensure the reports generated by the AI are accurate.

MCP configuration

No response

Chrome DevTools MCP version

latest

Chrome version

No response

Coding agent version

No response

Model version

No response

Chat log

No response

Node version

No response

Operating system

None

Extra checklist

• I want to provide a PR to fix this bug