How Cursor handles this
You are in the editor, you have just pushed the theme change, and the live site has an empty cache. In the agent chat you write which site to warm. Cursor calls list_sites, picks the handle, and calls start_preloader. The plugin starts crawling the sitemap in the background, so visitors find stored pages instead of waiting for rebuilds.
The tool call shows up in the chat as its own step with its arguments, so you can read exactly which site it names before it runs. Ask for progress and the next step is get_preloader_status, which reports whether a run is active and how far through the URL list it has got. Asked to stop, Cursor calls stop_preloader. Pages already warmed stay cached, so a stop only ends the crawl.
Run Modes, under Settings > Agents > Approvals & Execution, decide whether Cursor asks before it uses an MCP tool. Since Cursor 3.6 the recommended default is Auto-review, which runs allowlisted calls at once and sends the rest to a classifier, and Cursor says it is not a security boundary. Allowlist runs only what you listed and asks about everything else. Run Everything asks nothing. xSpeed Hub has no confirmation of its own, so a tool you allowlist, a call a classifier lets through, or any call under Run Everything starts a crawl as soon as your connection allows writes. Because the chat sits beside your files, you can also ask Cursor to note in the pull request which pages were warmed after the release, using what get_preloader_status returned.
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.
{
"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.
Customize > xspeedhub (on)
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.
I pushed the new header to production. Use xSpeed Hub to warm the shop site and show me the preloader status.
Using xSpeed Hub, is a preload still running on blog? Check, and stop it if it is.
With xSpeed Hub, warm all my sites, but list them first and confirm with me before you call anything.
What happens, step by step
The first visitor to an uncached page pays for building it. The preloader fixes that by visiting your pages first. Through xSpeed Hub an agent can start the crawl on one site or on every site in a workspace, watch how far it has got, and stop it. The crawl runs inside WordPress in the background; the agent only starts it, reads its progress and stops it.
01
Check the cache and the preloader
get_cache_status tells the agent whether page caching is on, because there is nothing to fill when it is off. get_preloader_status shows whether a crawl is already running, so the agent does not start a second one.
02
Make sure the preloader is switched on
The preloader module is off on a fresh install, and the site refuses start_preloader until it is on. The agent can read the module with get_settings and, if you agree, switch it on with update_settings on the preloader module. The Hub tells the agent to briefly confirm intent before an update_settings call, but that is guidance to the agent, not a lock.
03
Start the crawl
start_preloader makes the site read its sitemap, follow any nested sitemap indexes, drop URLs that your cache exclusion rules skip, and queue the rest. The call returns once the queue is built; the pages are then fetched in the background, a few per run. For every site at once the agent passes site: "all", and the Hub starts the sites one after another with a one-second pause.
04
Watch the progress
get_preloader_status reports whether a crawl is running, how many URLs it has processed out of the total, the last URL it visited and the most recent errors. It works on one site at a time, so for several sites the agent calls it once per site.
05
Read what failed
A page that answers 403 or 406 usually means a server firewall is blocking the warmer by its user-agent, not that the page is broken. The status message names the user-agent to allow. A crawl that queues no URLs is reported as an error with the reason, not as a success.
06
Stop it if the server struggles
stop_preloader ends a running crawl. It is safe mid-run: pages already warmed stay cached. The agent can stop one site or every site in the workspace.
Reference
| Start and stop | start_preloader and stop_preloader (write) |
|---|---|
| Progress | get_preloader_status (read): running or not, processed out of total, last URL, recent errors |
| What gets crawled | The sitemap (the WordPress one at /wp-sitemap.xml unless you set another), nested indexes followed, excluded URLs skipped |
| Needs | The preloader module switched on (it is off by default) and page caching on |
| Pages per run | 5 by default, 1 to 50, set in the preloader module |
| Schedule setting | manual, hourly, daily or weekly, set in the preloader module |
| Every site at once | start_preloader and stop_preloader accept site: "all"; get_preloader_status does not |
| Fan-out pacing | Sites start one after another with a 1 second pause, one result per site |
| Warmer user-agent | xSpeed-Warmer/1.0, the string a firewall rule would need to allow |
| Stopping | Safe mid-run; warmed pages stay cached |
| Read-only access | A Viewer member's sign-in, or the connection token with Read-only everywhere on: start_preloader and stop_preloader are refused; get_preloader_status works. An OAuth sign-in by anyone else gets the scopes the client asks for. |
| Confirmation | None on the Hub; your client's prompt is the gate |
Rules worth keeping
- Warm after a purge, not instead of one. The Hub tells the agent that starting the preloader after a purge saves the next visitor from paying for the rebuild.
- A crawl requests every page in the sitemap, so it adds load to the origin. Start with one site, and on a small host keep Pages per run low or stop the crawl if the site slows.
- start_preloader and stop_preloader are write tools. The Hub calls the preloader low-risk and only asks the agent to confirm when the target site is unclear or the call covers every site, so your client's approval prompt and read-only access (a Viewer sign-in, or the connection token with Read-only everywhere on) are the gates that matter.
- A fan-out only starts the crawls. Check each site with get_preloader_status afterwards, and treat some sites failing to start as a partial result, not a failure of the whole call.
Good to know with Cursor
Switch Run Mode to Allowlist (Settings > Agents > Approvals & Execution), list get_preloader_status and leave start_preloader and stop_preloader off the list, so each one asks. Under the default Auto-review, unlisted calls go to a classifier that Cursor says can allow something you would have blocked, and under Run Everything the two preloader writes run silently, on any site the agent names, and the chat only shows them afterwards. Run Modes apply to local agents: Cursor says Cloud Agents never ask for approval, so give a Cloud Agent only a read-only Hub connection, meaning the connection token with Read-only everywhere on (or a Viewer sign-in). A refusal from a site whose preloader module is off appears as a failed tool step in the chat, not as a started crawl.
More prompts for this job
They work in any client connected to xSpeed Hub.
Use xSpeed Hub to start the preloader on shop and tell me how many URLs it queued.
Using xSpeed Hub, how far through its crawl is blog? Show processed against total and any errors.
With xSpeed Hub, purge the cache on docs, then warm it again.
Use xSpeed Hub to start the preloader on every site in my workspace and tell me which ones refused.
Use xSpeed Hub to stop the preloader on shop. It is slowing the server down.
The preloader on shop says it is disabled. Use xSpeed Hub to turn the module on, then start the crawl.
Using xSpeed Hub, which of my sites has no running crawl and a low hit ratio?
Frequently asked questions
Keep going
Warm the cache with other agents
More with Cursor
Documentation
- How to warm your cache ahead of visitors
- How to crawl your sitemap
- How to enable Page Cache
- Managing your fleet with xSpeed Hub
- How to write prompts for xSpeed Hub
- Cursor + xSpeed
- Every AI agent that works with xSpeed
From the blog