MCP server for Lighthouse reports from PageSpeed Insights API
Single tool with complete schema and reasonable descriptions. Tool name follows verb_noun pattern (get-lighthouse-report). Input schema is well-defined with Zod validation, enums for category/strategy, and numeric bounds for timeout. However, output schema is not formally documented, response structure is inferred from code (text content only). Parameter descriptions are adequate (72 chars avg) but lack actionable guidance on when to use each category or strategy. No error recovery hints provided. Tool is read-only (safe), but error handling returns generic failure messages without suggesting next steps.
Get Lighthouse performance report for a URL
Output schema not formally documented. Response structure (text content only) is inferred from implementation, not declared. LLMs cannot plan downstream operations or extract structured data.
Error messages lack recovery guidance. 'Failed to retrieve Lighthouse data' and 'Error: <message>' do not tell the LLM what to do next (retry, check URL format, verify API key, etc.).
Parameter descriptions do not explain WHEN to use each category or strategy. E.g., 'Lighthouse category to include' does not clarify that performance measures speed, accessibility measures WCAG compliance, etc. LLMs cannot reason about which category to select.
API key exposed via environment variable without validation. If PAGESPEED_API_KEY is unset, empty string is used silently, causing API failures with no clear error message.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Timeout parameter default (60000 ms) differs from internal makePageSpeedRequest default (150000 ms). Inconsistency may confuse users about actual timeout behavior.