Read a cited page’s evidence
Useanalyze_page_performance with the project, run, and a URL drawn from the run’s source results. It returns data.chunks, data.statements, data.chunks_source, data.n_chunks, and data.n_statements when available.
The chunks describe captured page content. The statements are AI answer statements that cited the page in that measurement. An assistant can compare the two and explain which claims have clear support. That explanation is an assessment of retrieved evidence, not a new measured score.
If chunks_source is unavailable, say that captured page content is unavailable. The citing statements may still be useful, but they do not supply the missing page sections. The tool does not promise to crawl a fresh version of any URL on demand.
Read technical features
get_technical_features reads the selected page’s recorded features. Inspect matched_url because a domain or partial URL can match a page. The result includes feature keys, values, and sometimes an observed run_spread across pages.
Do not turn an observed range into a universal target. These data describe pages present in the run; they do not establish what separates cited pages from every uncited page. A null normalized_score is unavailable, not a zero quality score.
feature_group filters by the returned feature-key text on this route. The MCP argument feature_domain is not applied by the current HTTP route; omit it for this edition.
Read technical recommendations over HTTP
Configured HTTP integrations can request:url, impact, and effort. Use impact and effort labels returned in your data rather than inventing categories. Returned data.recommendations can contain url, domain, title, description, effort, impact, and implementation_steps. current_score and feature can be null.
data.n_recommendations counts returned recommendations. Despite its name, data.n_urls currently counts distinct domains in this response; do not label it “distinct page URLs.”
There is no equivalent named external MCP tool in the 43-tool catalog covered here. Content recommendations, briefs, and the Content Playground remain product workflows unless a separate supported integration is provided. Do not advertise content editing or publishing through these read tools.