Scan a website for speed problems with Grok Build

Scanning a site's speed with Grok Build means asking xAI's terminal agent to run an xSpeed Scan and return the graded report with the fixes ranked and a shareable link.

list_sitesget_speed_scanrun_speed_scan

Last updated: October 2026

Example session · Grok Build

You: Use the xSpeed Scan MCP to scan example-shop.com and tell me the biggest fix.

run_speed_scan(url: example-shop.com, strategy: both) read-only scan mcp
→ Desktop grade C, mobile grade D. Ranked fixes: page cache (every page rendered per visitor), then image size. Report link included.

Grok Build: Biggest fix: the server renders every page for each visitor, so a page cache is the top item. Mobile is the weaker run and image size is the second item. The report link has the full list.

You: Is that the lab grade or what real visitors see?

Grok Build: The grade is lab data. The report has a separate real-visitor strip, and it says if the site has too little Chrome traffic for field numbers.

How Grok Build handles this

You are in a terminal with Grok Build and you want a number for a site, not a hunch. Ask it to scan the URL. With the no-account xSpeed Scan MCP added, it calls run_speed_scan on any public address and returns a grade out of 100, the desktop run as the headline and the mobile run beside it, plus the failing checks with their evidence and a link to the full report. That is a read-only route, so it is safe to try first on a site that is not yours.

For a site in your own xSpeed Hub, the Hub route scans by handle and also exposes get_speed_scan, which reads the history of earlier reports for free. A short sequence works well: read the history, run one fresh scan after your change, and ask Grok Build which grade moved. If the two runs differ, mention both devices, since Google ranks on mobile.

Ask for one device when that is all you need. A desktop-only or mobile-only scan costs half the quota and makes that device the headline, which is handy when you are chasing a single failing check and want a quick re-test. Remember that the grade is lab data from one run: the report also shows what real Chrome visitors saw, and the two can disagree.

Set up Grok Build once

Already connected? Skip to the prompts. Alternatives and troubleshooting are on the Grok Build 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 with grok mcp add

Run this once. Add --scope project to write the server to .grok/config.toml in the current directory instead of your user config, which is only sensible if the repo never holds a token. Grok Build handles the OAuth flow for a remote server on its own.

Terminal

grok mcp add --transport http xspeedhub https://app.xspeedcache.com/xspeed/mcp

3Sign in from the TUI

Start grok and open /mcps. Select xspeedhub and press i to authenticate, or just use a Hub tool and let the browser flow start on first use. Sign in to xSpeed Hub and approve access. There is no token to paste; Grok Build stores the credentials it receives in ~/.grok/mcp_credentials.json. If the connection fails, run grok mcp doctor xspeedhub.

Text

/mcps

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 Grok Build 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 Grok Build 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. Grok Build starts in Ask mode, which prompts for anything a rule has not already allowed, so a Hub write tool prompts until you say otherwise. You can allow the reads with a rule such as MCPTool(xspeedhub__get_*) in the permission section of your config, and leave the writes unmatched. Evaluation runs deny, then ask, then allow. Two other modes remove the prompt: Always-approve, from /always-approve, Ctrl+O or --always-approve, approves tool calls unless a deny rule or a hook stops them, and Auto lets a classifier approve tools it judges safe. If a Hub write must never run unprompted, add a deny rule. The Hub itself has no confirmation step.

What do I ask?

Three prompts written for Grok Build. More for this job are below.

Prompt for Grok Build

With the xSpeed Scan MCP, grade [your site URL] and give me the desktop and mobile grades and the top three failing checks.

Prompt for Grok Build

Ask xSpeed Hub for a mobile-only scan on shop and tell me whether the images check passes now.

Prompt for Grok Build

Using xSpeed Hub, read the scan history for shop and tell me the last three grades.

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 Grok Build

Grok Build starts in Ask mode, which prompts for any tool call that no rule has allowed, so run_speed_scan prompts until you say otherwise. A rule such as MCPTool(xspeedhub__get_*) allows the history read and leaves the scan on the prompt. Always-approve and Auto mode remove that prompt, and deny rules still apply, so add one for any tool that must never run unprompted. On xSpeed Hub, run_speed_scan is a write tool because each scan spends the shared speed-test allowance, and a read-only connection (the connection token with Read-only everywhere on, or a Viewer sign-in) refuses it, which is the firm gate. The Scan MCP is read-only for every tool, so it is the better choice for a quick look at a site you do not run, within its limit of four scans per caller per ten minutes.

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

Yes, through the xSpeed Scan MCP. It needs no account, grades any public URL, and is read-only. The limit is four scans per caller per ten minutes, and the report it links is public.

It runs and grades only the mobile Lighthouse run, at half the quota, and mobile becomes the headline instead of desktop. The default scan runs both and grades desktop as the headline with mobile beside it.

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 Grok Build

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 Grok Build.

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

Open xSpeed Hub Grok Build setup