What actually goes wrong
Three failure modes, specific to this kind of site. Everything below is aimed at them.
Settings pages assume you’re an engineer
Defer which scripts? Exclude which cookies? Every caching plugin ships expert-level questions to non-expert users — and wrong answers look like a broken site.
Optimization is not one-and-done
Every plugin update, theme change, and new page can shift performance. Nobody re-tunes by hand every week — so sites quietly regress.
AI tools can advise but not act
ChatGPT can tell you what to change; without a real connection to the site, you are still the one clicking through settings pages.
How the agent connection works
A real MCP server in every install, a Hub that scales it to your whole fleet, and scopes that keep the keys in your hand.
Built-in MCP server
Every xSpeed Cache install exposes its own MCP endpoint. Claude, ChatGPT, or any MCP client connects directly — no middleware, no extra plugin.
MCP setup guidexSpeed Hub — one connection, every site
Link your AI to the Hub once and it reaches your whole fleet: benchmark all sites, fix the slow ones, report back — from a single conversation.
Connect the HubThe full tool surface
run_benchmark, update_settings, purge_cache, start_preloader, optimize_site and more — every CLI command is a tool your agent can call.
MCP tool referenceOne-command optimization
optimize_site runs the whole loop — benchmark, tune, warm, re-test — as a single call. "Make it fast" is now a literal instruction.
How it worksScoped connections
Per-site access levels, read-only credentials, denied tools. Give an agent exactly the power you mean to — and nothing more.
Scopes guideAI privacy & GDPR mode
One switch that every AI feature checks before recording visitor data. Agent-era convenience without a compliance surprise.
Privacy guide6 ways people put xSpeed to work
Each one shows when it applies and what to set up — or what to ask your AI for. Filter by role in the sidebar, or use the chips below.
“Make it fast” — literally
You know your site is slow. You do not want to learn what TTFB stands for to fix it.
WHEN
Any time. This is the front door of the whole product.
SETUP
Connect Claude or ChatGPT and say it. optimize_site runs benchmark → tune → warm → re-test as one loop and reports the score.
A weekly tune-up you never think about
Sites regress quietly — a new plugin here, a heavier page there. Nobody re-tunes by hand every week.
WHEN
On a schedule your assistant keeps, not one you keep.
SETUP
Ask your AI to benchmark weekly and fix regressions; score history shows the trend it is defending.
Purge and warm from your deploy pipeline
Every deploy leaves stale cache behind, and clicking “purge” after every release is a chore that gets skipped.
WHEN
CI finishes a deploy.
SETUP
Call purge_cache and start_preloader via MCP or WP-CLI from the pipeline — fresh cache, pre-warmed, no human step.
Start the AI in read-only mode
Letting software change production settings on day one is a reasonable thing to refuse.
WHEN
You want the diagnosis before you grant the treatment.
SETUP
Scoped connections support read-only credentials: the agent benchmarks and recommends; you click apply. Widen access when it earns it.
Triage thirty sites in one conversation
Fleet problems arrive all at once; fleet fixes usually do not.
WHEN
Several sites drift below target, or a fleet-wide audit is due.
SETUP
One Hub connection reaches every site: list them, benchmark them all, fix the stragglers, get one report back.
Adopt AI features without a GDPR surprise
Legal asks one question: what visitor data does the AI see?
WHEN
Before enabling AI features on an EU-facing site.
SETUP
AI privacy & GDPR mode is one switch every AI feature checks before recording visitor data — and it defaults to caution.
No scenarios under that role yet — pick “All” to see every one.
The exact prompts to use
Connect via xSpeed MCP or the Hub, then paste any of these into Claude or ChatGPT.
The one-liner
Using xSpeed, make this site score 100 on PageSpeed. Benchmark first, change whatever settings you need, warm the cache, re-test, and show me exactly what you changed.
The standing order
Every Monday, using xSpeed, benchmark my site. If the score dropped below 90, find out what changed, fix it, and summarise what you did. If everything is fine, just tell me the score.
The cautious start
Using xSpeed in read-only mode, benchmark my site and give me a prioritised list of what you would change and why — but change nothing yet. I’ll approve the list before you apply it.
Or just ask your AI
Every xSpeed Cache install speaks MCP. Connect Claude or ChatGPT once — directly or through xSpeed Hub — and everything on this page becomes something you ask for instead of something you configure.
you › This site scores 45. Make it fast.
xspeed · run_benchmark → 45 · LCP 4.1s
xspeed · page cache + preloader → on
xspeed · minify css/js · delay 3 scripts
xspeed · warmed 48 URLs from sitemap
xspeed · run_benchmark → 100 · LCP 0.9s
Done — 100/100, and your assistant knows how to keep it there.
See the story for your business
Each one is a real scenario — the moment speed bites, the fix, and the exact prompt to hand your AI.
Frequently asked questions
Keep reading
← Previous playbook
Membership & LMS
Object caching for the logged-in side, full-page speed for the public side, and precise rules for the line between them.
Next playbook →
WooCommerce stores
Static product pages in milliseconds, a cart that stays live, and a database that stays lean — store-aware caching out of the box.