Scan a website for speed problems with Cursor

Scanning a site's speed with Cursor means asking its agent, from the editor where you work on the theme or plugin, to run an xSpeed Scan and show the graded result beside your code.

list_sitesget_speed_scanrun_speed_scan

Last updated: October 2026

Example session · Cursor

You: The new landing template feels slow. Use xSpeed Hub to scan shop and find the cause in this repo.

run_speed_scan(site: shop, fresh: true, strategy: both) uses quota
→ Desktop grade B, mobile grade D. Top failing check: LCP image discoverable early. The hero image is large and lazy-loaded, so the browser finds it late.

Cursor: Mobile is the weak run. The scan says the hero image is large and lazy-loaded, so it is found late. I can find it in the landing template, serve a smaller size, preload it and keep it out of lazy loading. Want me to edit the template?

You: Yes, then scan again.

How Cursor handles this

You are in a theme repository and the page you just reworked feels heavier. Instead of switching to a browser, you ask the Cursor agent to scan the site. It calls run_speed_scan through xSpeed Hub with the site handle, waits up to about 50 seconds, and returns the grade, the failing checks and a report link. Because it sits next to your files, the agent can connect a finding to the code that causes it: a render-blocking stylesheet, an oversized hero image, a template that skips the cache.

The scan grades the desktop run as the headline and the mobile run beside it, on one 0 to 100 rubric. Every failing check carries its evidence and the xSpeed feature that would fix it. The report also has a Markdown twin at the same address with .md added, which suits pasting the full diagnosis into the chat.

For a competitor's site or a client's site outside your Hub, connect the no-account Scan MCP and ask for a URL instead of a handle. The two servers sit side by side in Customize, and you can switch either one off when you do not need it.

Set up Cursor once

Already connected? Skip to the prompts. Alternatives and troubleshooting are on the Cursor guide.

1Put your sites in xSpeed Hub

Sign in at app.xspeedcache.com with Google or email; the Hub is free and has no site cap. Then connect each WordPress site from its own dashboard: click Connect Hub in the xSpeed Cache top bar, then Connect via xSpeed Hub. Each site needs the free xSpeed Cache plugin.

2Add xSpeed Hub to mcp.json

Add the server to ~/.cursor/mcp.json to have it in every project, or to .cursor/mcp.json inside a project to scope it there. The entry is just a name and a url.

~/.cursor/mcp.json

{
  "mcpServers": {
    "xspeedhub": {
      "url": "https://app.xspeedcache.com/xspeed/mcp"
    }
  }
}

3Enable it and sign in

Open Customize in the sidebar, find xspeedhub and make sure it is switched on. When Cursor starts the OAuth sign-in, approve access on the xSpeed Hub page. There is no token to paste. If the server shows an error, open the Output panel with Cmd+Shift+U and pick MCP Logs.

Text

Customize  >  xspeedhub  (on)

4Add the xSpeed Scan MCP for sites you do not run

Prompts that scan a public URL use the xSpeed Scan MCP, a separate server from the Hub. Add it to Cursor the same way as the Hub above, as a second server (xspeedscan, for example) with the URL https://xspeedcache.com/scan/mcp, and leave out any sign-in setting, such as an OAuth or auth option or an Authorization header, because the Scan MCP needs no account, token or sign-in. On a plan that allows one custom connector, add whichever server the prompts you use need.

Text

https://xspeedcache.com/scan/mcp

1 / 4

Full Cursor setup, sign-in options and FAQ

Before you send a prompt that changes something

Reads change nothing on your sites, though contact_support emails xSpeed support. Writes run once your connection allows them, so your client's approval prompt and a read-only connection (the token's Read-only everywhere switch, or a Viewer sign-in) are the gates that matter. Cursor decides how to treat an MCP tool call by its Run Mode, and you can expand a call to see its arguments. MCP follows the same Run Modes as terminal commands, set under Settings > Agents > Approvals & Execution. The recommended default, Auto-review, runs allowlisted calls at once and sends the rest to a classifier, not to you. Allowlist runs only what you listed, and with an empty list it asks you every time. Run Everything runs every tool call without asking. xSpeed Hub has no confirmation step of its own, so a tool you allowlist, or any call under Run Everything, writes as soon as your connection allows writes. Auto-review is a model, and Cursor says it is not a security boundary.

What do I ask?

Three prompts written for Cursor. More for this job are below.

Prompt for Cursor

Use xSpeed Hub to scan the site this repo deploys to and tell me which failing checks I can fix in this codebase.

Prompt for Cursor

Here is a scan report link. Use the xSpeed Scan MCP to read it, then open the files that cause the top two failing checks.

Prompt for Cursor

With xSpeed Hub, run a fresh mobile-only scan after my last commit and compare it with the previous grade.

What happens, step by step

An xSpeed Scan grades a site from 0 to 100 using server probes (time to first byte, compression, cache evidence, redirects, platform detection) plus a Lighthouse run, and names the fix for every failing check. An agent can start one two ways. Through xSpeed Hub, run_speed_scan scans a site you have connected. Through the no-account Scan MCP at https://xspeedcache.com/scan/mcp, it scans any public URL and cannot change anything. Either way you get a report link to share.

Check what the site already scored

get_speed_scan without a scan id returns the site's scan history, newest first, plus a summary of the latest report. It starts nothing and spends nothing, so the Hub tells the agent to read it first when you ask what a site scored.

Start a scan

run_speed_scan runs the full scan on a site you connected to the Hub. strategy is both (the default), desktop or mobile; one device alone costs half the engine's quota. The scan takes 20 to 60 seconds. The Hub waits about 50 seconds and, if the scan is still running, hands back a scan id.

Poll if it is still running

Call get_speed_scan with scan_id. While the scan runs it returns the running status and a note to ask again in about 20 seconds. The report page is live from the start and refreshes itself.

Read the result

The summary carries the grade, the four dimension scores, what was measured and the top fixes ranked by points recoverable, each with the xSpeed feature that fixes it. The headline grade is the desktop run. The mobile run is graded alongside it, so the agent should mention both when they differ.

Scan a site you do not run

For any other public URL, the agent uses the Scan MCP at https://xspeedcache.com/scan/mcp. There is no account and no sign-in. It takes a url, returns the same graded report and a link at xspeedcache.com/scan/r/<id>, and is limited to 4 scans per caller per 10 minutes.

Fix, wait, re-scan

After a change, wait several minutes and scan again with fresh: true. Without it, the scan service reuses any report from the last 10 minutes and the agent will show you the old number.

Reference

Hub route	run_speed_scan, then get_speed_scan with scan_id; only for sites connected to your Hub
Scan MCP route	https://xspeedcache.com/scan/mcp, no account, 5 tools, read-only, any public URL
Scan MCP limit	4 scans per caller per 10 minutes
Why run_speed_scan is a write	It spends a shared scan allowance and stores a report; it changes nothing on the site
Read-only Hub connection	run_speed_scan is refused; get_speed_scan still works. Applies to the connection token with Read-only everywhere on and to a Viewer sign-in; an OAuth sign-in gets the scopes the client asks for
Duration	20 to 60 seconds; the Hub waits about 50 seconds, then returns a scan id
Poll argument	scan_id on the Hub tool, scanId on the Scan MCP tool
strategy	both (default), desktop or mobile; one device costs half the quota and becomes the headline
fresh	true forces a new measurement; otherwise a report from the last 10 minutes is reused
Report link	xspeedcache.com/scan/r/<id>; add .md for a Markdown twin an agent can read
Report visibility	Hub scans are filed unlisted; Scan MCP scans are public in the scan directory by default

Rules worth keeping

Read before you spend. Ask for the scan history first; start a new scan only when you want a fresh measurement or something changed.
A report link is not private. An unlisted report is kept out of the scan directory and search, but anyone with the URL can open it. A scan started through the Scan MCP is public by default.
run_speed_scan changes nothing on your site, but the Hub classes it a write, so a read-only connection (the connection token with Read-only everywhere on, or a Viewer sign-in) refuses it and your client may prompt for it. Use get_speed_scan, or the Scan MCP for a public URL, when you want a strictly read-only route.
Compare like with like. A scan that lost its Lighthouse run is marked partial and is not comparable with a full one, and the lab grade is not the same as real-visitor Core Web Vitals, which the report shows separately.
A scan cannot reach private, local or password-protected sites, and a site behind a bot challenge returns a report that says its server checks were not measured.

Good to know with Cursor

Since Cursor 3.6 the default Run Mode is Auto-review, which sends MCP calls that are not allowlisted to a classifier, not to you, and Cursor says that classifier is not a security boundary. Run Everything asks nothing. In either mode run_speed_scan on the Hub runs as soon as the agent decides to, and an agent told to keep trying until the grade is an A can start many fresh scans, each spending the shared speed-test allowance. Switch Run Mode to Allowlist (Settings > Agents > Approvals & Execution), list get_speed_scan, and leave run_speed_scan off so it asks. Give the agent a clear stopping rule, such as one scan before the change and one after. Cloud Agents never ask, so give one a read-only connection: the connection token with Read-only everywhere on. Scans can only see what is deployed, so uncommitted edits in the editor do not change the grade until they are live.

More prompts for this job

Prompts about your own sites work in any client connected to xSpeed Hub. A prompt that scans a site you do not run, or names the scan server, needs the xSpeed Scan MCP from the setup above.

Prompt

Using xSpeed Hub, what did the shop site score on its last speed scan? Do not start a new one.

Prompt

Use xSpeed Hub to run a speed scan on blog and give me the three fixes that recover the most points, plus the report link.

Prompt

With xSpeed Hub, scan only the mobile view of shop and tell me the grade.

Prompt

I changed the theme this morning. Use xSpeed Hub to run a fresh scan on docs and compare it with the previous report.

Prompt

Use the xSpeed Scan MCP to grade [competitor URL] and tell me what its biggest weaknesses are.

Prompt

The speed scan on shop is still running. Ask xSpeed Hub to check it again and tell me when it finishes.

Frequently asked questions

No. A scan measures the live public page. Your edits show up in a scan only after they are deployed to the URL being scanned, so deploy first and then ask Cursor for a fresh scan.

Allowlist is the most predictable. List get_speed_scan, which reads history and starts nothing, and leave run_speed_scan off the list so each scan asks first. Auto-review, the default, is a classifier that decides for calls you did not allowlist, and Cursor says it is not a security boundary.

run_speed_scan runs the full xSpeed Scan: server probes plus Lighthouse, a 0 to 100 grade across four dimensions, the fix for every failing check and a shareable report link. run_speed_test returns a raw PageSpeed Insights score with Core Web Vitals and no graded report.

Yes, through the xSpeed Scan MCP at https://xspeedcache.com/scan/mcp. It needs no account and no sign-in, accepts any public URL and is read-only. The Hub route only scans sites you connected to your own workspace.

The Scan MCP allows 4 scans per caller per 10 minutes. Scanning the same URL again within 10 minutes returns the existing report unless the agent passes fresh: true. A Hub scan spends a shared allowance of the scan service, which is why the Hub treats it as a write tool.

Reports started from the Hub are filed as unlisted, which keeps them out of the public scan directory and search results, but the link itself still opens for anyone who has it. Reports started through the Scan MCP are public by default.

The headline grade is the desktop run, because it is the one you can reproduce in your own browser. The mobile run is graded on the same rubric and returned alongside it. Google ranks on mobile, so when the two differ the agent should tell you both.

Keep going

Scan a site's speed with other agents

More with Cursor

Purge the cache Find out why pages are not cached Track PageSpeed scores Raise the PageSpeed score Tune cache settings Warm the cache Set up Redis object cache Manage Cloudflare Manage every site at once

Documentation

From the blog

Scan a site's speed, from Cursor.

xSpeed Hub is free and has no site cap. Connect your sites once and ask from the agent you already use.

Open xSpeed Hub Cursor setup