How to Audit Core Web Vitals From Claude with PageSpeed MCP (2027)

Audit Core Web Vitals from Claude using Otto's PageSpeed MCP server without a Google API key, then stage fixes as WordPress drafts or GitHub pull requests.

As seen in

  • Fox News
  • ABC News
  • CBS
  • NBC
  • Associated Press
  • USA Today
  • Bloomberg
  • HuffPost
Otto the beaver next to a browser window icon
On this page6

Audit Core Web Vitals from Claude by connecting Otto's hosted MCP server (connect Claude to Otto MCP) with one otto_mcp__ token and running the built-in psi_analyze_url toolβ€”no user Google API key needed. Claude pulls trailing 28-day CrUX field data and Lighthouse lab diagnostics; use Step 2 for official Core Web Vitals pass/fail thresholds at the 75th percentile. Fixes stage as WordPress drafts (29 tools) or GitHub pull requests with Vercel previews for your approval.

Copy-paste Claude prompts to audit Core Web Vitals and stage fixes (start here)

Paste these five prompts directly into Claude or Cursor once your Otto MCP connection is active.

1. Mobile and desktop audit with CrUX vs. Lighthouse separation

Run psi_analyze_url for https://example.com/ on both mobile and desktop. 
Separate the output into two distinct sections:
1. Field Data: Report the trailing 28-day Chrome User Experience Report (CrUX) metrics at the 75th percentile: Largest Contentful Paint (LCP), Interaction to Next Paint (INP), Cumulative Layout Shift (CLS), First Contentful Paint (FCP), and Time to First Byte (TTFB). State clearly whether the page passes or fails the Core Web Vitals assessment.
2. Lab Diagnostics: Report the Lighthouse lab performance score, Speed Index, Total Blocking Time (TBT), and the top three resource savings opportunities (unoptimized images, render-blocking stylesheets, or unused JavaScript).

2. Paid ad landing-page triage with spend and conversion cross-referencing

Audit our top landing pages currently receiving paid traffic.
First, call meta_get_campaigns and google_ads_get_campaigns to identify the active landing page URLs with the highest click volume and spend over the last 30 days.
Next, run psi_analyze_url on each URL using the mobile strategy.
Cross-reference the mobile LCP, mobile INP, and Lighthouse lab scores against ga4_run_report conversion rates for each page.
Flag any landing page spending more than $500 where mobile LCP exceeds 2.5 seconds or INP exceeds 200 milliseconds, and list the exact script or asset causing the bottleneck.

3. Google Search Console high-impression organic audit

Call gsc_search_analytics to pull our top 15 pages ranked by organic search impressions over the trailing 28 days.
Run psi_analyze_url on each URL.
Group the pages into three buckets:
- Good: LCP <= 2.5s, INP <= 200ms, CLS <= 0.1
- Needs Improvement: LCP between 2.5s and 4.0s, INP between 200ms and 500ms, or CLS between 0.1 and 0.25
- Poor: LCP > 4.0s, INP > 500ms, or CLS > 0.25
For pages in the Poor bucket that rank for high-intent search terms, output the specific DOM elements causing layout shifts and the primary LCP image element.

4. Stage WordPress Core Web Vitals fixes as drafts

Review the psi_analyze_url diagnostics for our WordPress page at https://example.com/services.
Using the wordpress_get_page and wordpress_update_page tools:
1. Inspect the current page content and identify any large hero images lacking explicit width and height attributes or missing lazy-loading exclusions.
2. Draft an updated version of the page content that adds fetchpriority="high" to the above-the-fold hero image and sets explicit aspect ratios to resolve Cumulative Layout Shift.
3. Save the update strictly with status="draft" so the existing live page remains untouched until approved.

5. Open a GitHub pull request with a Vercel preview for code-managed sites

We have a slow Largest Contentful Paint (3.4s) on our Next.js marketing site.
Inspect the repository files for the landing page hero section.
Generate a git branch and pull request that:
1. Preloads the critical above-the-fold hero image using next/image with priority={true}.
2. Adjusts web font loading to use font-display: swap.
3. Preconnects to third-party script origins.
Provide the GitHub pull request link and confirm that Vercel has generated an isolated deployment preview URL for inspection before merging.

Manual PageSpeed tabs vs. DIY local PageSpeed MCP wrappers vs. Claude + Otto MCP

Auditing Core Web Vitals usually forces a compromise between repetitive manual browser tests and fragile local developer setups. The table below compares manual testing, standalone open-source MCP wrappers, and Otto MCP's 227 tools across 15 integrations.

CapabilityManual PageSpeed UI & CSVsDIY Local stdio PageSpeed MCPClaude + Otto MCP (https://otto.islaintel.com/api/mcp)
Setup & MaintenanceNo setup; paste URLs manually into pagespeed.web.devNode.js install, local stdio configs, per-device troubleshootingHosted remote HTTP MCP; connect once with an otto_mcp__ token
API Key RequirementNone (interactive browser limits apply)Requires provisioning a Google Cloud project and PageSpeed API keyZero user Google API key required; runs on Otto's platform key
Cross-Tool TriageDisconnected; manual switching between Search Console, GA4, and AdsNone; single-purpose wrapper with no access to ads or CMS platformsIntegrated across Google Ads, Meta Ads, GA4, Search Console, and WordPress
Context Window Efficiency40,000 to 80,000 tokens per 3,000-row CSV uploadReturns full Lighthouse JSON payloads per runQuerying aggregated ad metrics through Otto MCP cuts token use 80–90% versus uploading a 3,000-row Ads Manager CSV; PageSpeed calls return structured tool output
Landing-Page RemediationRead-only; write tickets for developers or edit CMS by handRead-only; cannot modify theme code, pages, or imagesStages fixes as WordPress drafts (29 tools) or GitHub pull requests with Vercel previews
Approval GuardrailsManual human entryNo execution capabilitiesDraft-first staging; ad changes stay paused; changes wait for approval

Connect Claude to Otto MCP

Step 1: Connect Claude (or Cursor) to Otto MCP with zero Google API keys

Remote MCP servers are hosted on the internet and accessible from any internet-connected MCP client without local per-device installation or configuration, as documented in the Model Context Protocol remote servers specification. Otto implements this remote architecture over HTTP at https://otto.islaintel.com/api/mcp.

Users do not paste GA4 keys, Meta tokens, or WordPress app passwords into Claude. You connect your platforms once inside the Otto dashboard using standard OAuth. Your AI client holds only an organization-scoped bearer token prefixed with otto_mcp__, while all OAuth secrets and underlying API credentials stay on Otto's secure server.

Because Otto MCP includes the built-in psi_analyze_url tool powered by Otto's platform key, you do not need to create a Google Cloud project or provision a personal PageSpeed Insights API key. Follow the How to Connect Claude to Otto MCP setup guide or read the Connect PageSpeed Insights to Otto documentation to verify the connection.

Connecting Claude Desktop or Claude on the web

Add the Otto remote server to your Claude configuration:

{
  "mcpServers": {
    "otto": {
      "url": "https://otto.islaintel.com/api/mcp",
      "headers": {
        "Authorization": "Bearer otto_mcp_YOUR_TOKEN_HERE"
      }
    }
  }
}

For web-based Claude clients supporting Custom Connectors, enter https://otto.islaintel.com/api/mcp as the server URL and supply your bearer token. Supported clients include Claude, ChatGPT, Cursor, Claude Code, OpenClaw, and Hermes.

Connecting Cursor for agency and developer workflows

If your team manages code directly in Cursor, follow Connect Cursor to Otto MCP. Add the server under Cursor Settings > Features > MCP Servers:

  • Name: otto
  • Type: sse or http
  • URL: https://otto.islaintel.com/api/mcp
  • Headers: Authorization: Bearer otto_mcp_YOUR_TOKEN_HERE

Once added, Claude and Cursor gain access to all 227 Otto tools, including psi_analyze_url.

Step 2: Read your psi_analyze_url auditβ€”28-day CrUX field data vs. Lighthouse lab scores

According to Google for Developers, real-user experience data in PageSpeed Insights is powered by the Chrome User Experience Report (CrUX) dataset, which reports real users' First Contentful Paint (FCP), Interaction to Next Paint (INP), Largest Contentful Paint (LCP), and Cumulative Layout Shift (CLS).

When running psi_analyze_url, Claude returns two distinct categories of data: field data and lab data. Understanding the difference prevents wasted engineering effort.

       β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
       β”‚                   psi_analyze_url Call                      β”‚
       β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                      β”‚
            β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”΄β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
            β–Ό                                                   β–Ό
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”   β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚       CrUX Field Data         β”‚   β”‚         Lighthouse Lab Data           β”‚
β”‚  - Real Chrome user visits    β”‚   β”‚  - Simulated Moto G4 / Desktop run    β”‚
β”‚  - Trailing 28-day window     β”‚   β”‚  - Synthetic network throttling       β”‚
β”‚  - Evaluated at 75th %ile     β”‚   β”‚  - Instant point-in-time check        β”‚
β”‚  - Determines CWV pass/fail   β”‚   β”‚  - Uncovers diagnostics & code issues β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜   β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Per Google Chrome Developers (web.dev), Core Web Vitals thresholds require Largest Contentful Paint (LCP) within 2.5 seconds, Interaction to Next Paint (INP) of 200 milliseconds or less, and Cumulative Layout Shift (CLS) of 0.1 or less at the 75th percentile of page loads.

MetricGood (Passes)Needs ImprovementPoorAssessment Type
Largest Contentful Paint (LCP)≀ 2.5 seconds2.5s – 4.0s> 4.0sCore Web Vital
Interaction to Next Paint (INP)≀ 200 milliseconds200ms – 500ms> 500msCore Web Vital
Cumulative Layout Shift (CLS)≀ 0.10.1 – 0.25> 0.25Core Web Vital
First Contentful Paint (FCP)≀ 1.8 seconds1.8s – 3.0s> 3.0sDiagnostic Field Metric
Time to First Byte (TTFB)≀ 0.8 seconds0.8s – 1.8s> 1.8sDiagnostic Field Metric

CrUX metrics aggregate anonymized page loads over a trailing 28-day window. On low-traffic URLs lacking sufficient individual visits, PageSpeed Insights automatically falls back to origin-level CrUX metrics representing all pages across your domain.

Lighthouse lab data, by contrast, is a synthetic simulation run against a simulated Moto G4 device on an emulated throttled network connection. A page can score 95 in Lighthouse lab tests yet fail Core Web Vitals if real-world mobile visitors encounter long main-thread delays on complex scripts.

As noted by Google Search Central, achieving good Core Web Vitals is recommended because these metrics align with Google's core ranking systems for page experience. However, Otto never guarantees search rankings, Google Map Pack placement, or ChatGPT mentions; he structures websites, schema, listings, and technical performance so search engines and AI assistants can read and understand the business. Pair technical audits with the Otto AI SEO Agent to track rankings and search health over time.

Step 3: Triage slow ad and SEO landing pages and stage the fix as a WordPress draft or GitHub PR

Unlike ad-only tools that stop after flagging high cost per click, Otto allows you to diagnose the slow page and stage the technical fix in the same conversation. Review how Core Web Vitals drift leaks landing-page revenue to evaluate the conversion cost of slow mobile landing pages.

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ psi_analyze_url         │──► Identifies slow LCP image (3.8s) & high INP (340ms)
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
            β”‚
            β–Ό
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ Cross-Tool Triage       │──► Matches URL with Meta Ads spend ($1,200) & low GA4 conv.
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
            β”‚
            β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
            β–Ό                                         β–Ό
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ WordPress Draft Remediation           β”‚ β”‚ GitHub PR + Vercel Preview            β”‚
β”‚ - Uses 29 WordPress tools             β”‚ β”‚ - Clones branch, edits hero component β”‚
β”‚ - Applies fetchpriority="high"        β”‚ β”‚ - Preloads fonts & optimizes styles   β”‚
β”‚ - status="draft" protects live site   β”‚ β”‚ - Generates isolated preview URL      β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                    β”‚                                         β”‚
                    β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                                         β–Ό
                        β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
                        β”‚ Human Approval Gate             β”‚
                        β”‚ - One-tap WhatsApp confirmation β”‚
                        β”‚ - Nothing deploys without you   β”‚
                        β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Staging fixes on WordPress with draft protection

Under the WordPress REST API schema, creating or updating a page accepts a status parameter set to publish, future, draft, pending, or private. Otto MCP's WordPress integration provides 29 tools (15 read tools and 14 write tools) across posts, pages, media, categories, tags, comments, taxonomies, and search. In Otto, new posts and pages default to draft status.

For example, say you run a WordPress clinic landing page and psi_analyze_url reports an LCP of 3.8 seconds caused by an oversized hero banner. Claude uses wordpress_get_page to inspect the content. Claude then calls wordpress_update_page to replace the unoptimized image tag with an optimized WebP source, add explicit image dimensions to eliminate CLS, and set status: "draft". The existing live page continues serving visitors while your team reviews the draft in wp-admin.

Opening a GitHub pull request with a Vercel preview

For code-managed web applications, Otto integrates with GitHub and Vercel. For website updates covering pages, posts, hours, and photos, every code change on GitHub opens as a pull request with an isolated preview deployment.

Claude inspects your repository, updates the component to preload fonts or adjust script execution, commits the change to a new branch, and opens a pull request. Vercel automatically deploys a preview URL, allowing your team to test the layout and re-run psi_analyze_url on the staging deployment before merging.

Built-in automation: image compression and WhatsApp approvals

Otto handles ongoing site hygiene automatically:

  • Automatic Image Optimization: On June 15, 2026 (v1.1.5), Otto shipped Otto v1.1.5 automatic image optimization, compressing newly uploaded images automatically and re-processing existing site images during scheduled maintenance passes.
  • One-Tap WhatsApp Approvals: On June 2, 2026 (v1.1.0), Otto shipped enhanced WhatsApp integration, allowing business owners to approve or reject content changes directly from WhatsApp without logging into a web dashboard.
  • Automated SEO Tracking: On June 24, 2026 (v1.2.0), Otto shipped automated weekly SEO reports that track keyword movements and technical issues over time.

Enrique Gomez, Owner of Enzo Maverick Pilates and Experiences Sensory, said: "Otto has been pivotal for us. The ease of making website changes on the fly lets us sharpen our brand and revise as fluidly as we need." Other businesses running on Otto include Sensory Pilates & Wellness, Wyn Wellness, Complex Physical Therapy & Performance, Urban Boxing, and Tapper Training.

Explore the Otto AI Website Manager to see how code staging and draft management operate together.

Permission guardrails, Otto vs. ad-only tools, pricing, and FAQs

Connecting an AI assistant to live infrastructure requires strict security boundaries. Otto enforces structural limits across every connected tool.

Permission guardrails

Otto separates read-only analysis from approval-gated write operations:

  • Strictly Read-Only Integrations: Google Analytics 4, Google Search Console, Google Ads, Google Business Profile, YouTube, Google Sheets, Shopify, and PageSpeed Insights are read-only in Otto MCP, with zero ability to alter configurations or spend money.
  • Approval-Gated Staging: On GitHub, site updates open as pull requests with previews. On WordPress, updates default to draft.
  • Protected Core Pages: Checkout, contact, and legal pages always require explicit human approval in Otto core.
  • Audit Logging: Every automated deploy and site-edit run is recorded in otto_get_change_log.
  • Social & Ad Write Rules: LinkedIn Ads campaigns are created as DRAFT by default, and budget activations require confirm=true and reviewer or admin roles. Meta Ads write actions similarly require reviewer or admin status. Unlike WordPress drafts, publishing to Facebook Pages and Instagram through Otto MCP publishes live immediately and requires reviewer or admin credentials.

Otto Webmaster vs. ad-only tools (Opteo, Synter, Ryze AI)

CapabilityOpteoSynterRyze AIOtto Webmaster
Supported Ad NetworksGoogle Ads only40+ ad networksGoogle & Meta AdsMeta, Google, LinkedIn, X, ChatGPT Ads
Core Web Vitals & PageSpeedNoneNoneBasic SEO auditBuilt-in psi_analyze_url via platform key
Landing-Page RemediationNone; ad suggestions onlyNone; ad connectors onlyLimited; ad/GEO focusStages WordPress drafts & GitHub pull requests
Model Context ProtocolNo MCP serverMCP ad connectorProprietary chatHosted HTTP MCP (227 tools, 15 integrations)
Execution SafetySuggestions onlyClient-managed rulesAutomated changesHuman approval required for all live deploys

Unlike Opteo (which monitors Google Ads only with no SEO, website management, or MCP server), Synter (which connects 40+ ad networks but does not touch website code or CMS platforms), or Ryze AI (which focuses on ads and GEO without direct GitHub or WordPress page staging), Otto diagnoses ad efficiency and landing-page speed together, then stages the actual fix.

Honest limitations and pricing

Otto is built for service businesses and lean teams that need routine webmaster, SEO, and ad execution without agency retainers. It is not an agency white-label portal, and white-label reporting options are limited. Furthermore, the Otto terminal CLI is coming soon and is not live today; remote MCP is the primary AI connection. Otto does not support unannounced third-party platforms such as Wix, Webflow, or TikTok.

Review Otto Trial, Pro, and Master pricing:

  • $1 Trial (3 days): A 3-day trial of Pro covering website updates, SEO, and read-only ad and social access. Includes a custom homepage redesign built from your photos and copy, approved by you before publishing.
  • Pro Plan ($199/mo, or $99/mo billed annually): Full website management, technical SEO, and read-only access to ads and socials. Cancel anytime.
  • Master Plan ($3,500/mo, or $3,000/mo billed annually): All Pro features plus approval-gated write execution across paid ads and organic socials.

Otto MCP access uses your existing Otto subscription plan with no extra API surcharges.

Frequently asked questions

Can Claude alter my live website without my permission?

No. Every website modification staged through Otto MCP requires your approval before going live. WordPress pages and posts are saved with status: draft, while code edits open as GitHub pull requests accompanied by isolated deployment previews.

Do I need to create a Google Cloud project to use psi_analyze_url?

No. Otto MCP runs psi_analyze_url using Otto's verified platform credentials. You do not need a Google Cloud account, a billing profile, or a personal PageSpeed Insights API key.

Does fixing Core Web Vitals guarantee top Google rankings?

No. Google evaluates page experience alongside hundreds of content and relevance signals, and Otto never guarantees search rankings or Google Map Pack positioning. Otto structures websites, schema, listings, and reviews so search engines and AI assistants can read and understand the businessβ€”it does not promise compliance with every Google technical requirement.

Which platforms can Otto update directly?

Otto natively updates WordPress sites using 29 REST API tools and code-managed sites hosted on GitHub and Vercel. Unannounced platforms like Wix, Webflow, or TikTok are not supported.

Connect Claude to Otto MCP

Written by

Julius Washington

Founder & CEO at IslaIntel

Founder of IslaIntel and Otto, working to close the Caribbean's digital gap with practical AI.