How Gemini CLI handles this
You work in a terminal, you run more WordPress sites than you want to open one by one, and Gemini CLI is the agent you reach for. A typical request is "warm the cache on all my sites after the release". Gemini CLI calls start_preloader on the xspeedhub server with site set to "all", and the Hub runs the sites in sequence, a second apart. For fifteen sites that is about fifteen seconds before you see the answer.
The answer is one result per site, each marked ok or not, with data or an error. Gemini CLI can print it as a list, so a partial failure is easy to spot: fourteen warming, one refused because its preloader module is off. The seven fan-out tools are purge_cache, toggle_cache, update_settings, start_preloader, stop_preloader, optimize_site and run_benchmark. Any other tool takes one site per call, and Gemini CLI loops over them itself.
By default Gemini CLI asks before a tool call runs, unless the server is trusted or you start it with --approval-mode=yolo (or --yolo), which approves every call. That default is what you want for a write to every site. For a purge, toggle, settings change or preloader call, the Hub also tells the agent to confirm once, with the site count. The Hub does not enforce that, and it does not name optimize_site. Reads change nothing on your sites, and the connection's scope is the hard limit on writes, so the connection token with Read-only everywhere on, or a Viewer sign-in, refuses them whatever the CLI is set to.
Set up Gemini CLI once
Already connected? Skip to the prompts. Alternatives and troubleshooting are on the Gemini CLI 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 Gemini CLI
Run this once. --scope user writes the server to ~/.gemini/settings.json so it works in every project; without it Gemini CLI saves to the project's .gemini/settings.json, which is the default.
gemini mcp add --scope user --transport http xspeedhub https://app.xspeedcache.com/xspeed/mcp
3Authenticate with /mcp auth
Start gemini and run /mcp auth xspeedhub. A browser page opens where you sign in to xSpeed Hub and approve access, and Gemini CLI receives the redirect on a local port. There is no token to paste, and the client keeps the credentials it receives. Run /mcp afterwards to check that xspeedhub shows as connected and lists its tools.
/mcp auth xspeedhub
Full Gemini CLI 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. Gemini CLI shows a confirmation for an MCP tool call unless the server entry sets trust to true or you start Gemini CLI with --approval-mode=yolo (or --yolo), which approves every tool call. You can proceed once, always allow that tool, always allow the whole xspeedhub server, or cancel; the last two choices stop later prompts for those tools. Policy rules in ~/.gemini/policies can allow, deny or ask for a named tool by server and tool name. The Hub has no confirmation step of its own, so a trusted server, YOLO mode or an allow rule lets a write run as soon as your connection allows writes.
What do I ask?
Three prompts written for Gemini CLI. More for this job are below.
Use xSpeed Hub to start the preloader on all sites and print the status of each one.
With xSpeed Hub, run the benchmark on every site and rank them by how much faster the cached response is.
Ask xSpeed Hub to purge all sites, then list any that returned an error, with the error text.
What happens, step by step
If you run many WordPress sites, repeating one action on each is the slow part. Through xSpeed Hub, seven tools accept every site in a workspace, or a list of handles, in a single call. The Hub runs the sites one after another and returns a result for each, so a site that fails is named instead of hidden. Everything else in the Hub works on one site per call.
01
See what you can reach
list_sites returns each site's handle, URL, status and scopes. When you have more than one workspace, list_workspaces returns each one's handle, id, your role, its site count, which one is the default, and whether this connection can reach it.
02
Choose the scope of the call
The agent passes site: "all" for every site in one workspace, or an array of handles for exactly those sites. For a workspace other than the default it also passes the workspace argument. Omit workspace and the call acts on the default workspace.
03
Confirm once
Before a purge, toggle, settings change or preloader call that touches every site, the Hub tells the agent to say "this will touch all N sites" and wait for you, then make one call. That instruction does not name optimize_site, so for it your client's prompt is the gate. It is guidance to the agent, not a lock. It is one tool call, so one approval in your client covers every site in it.
04
Run it as one call
The Hub runs the sites one after another with a one-second pause between live calls, so a fleet purge is not a burst of traffic at your servers. Revoked sites are skipped from "all". Each site's action lands in the activity log as its own row.
05
Read the result per site
The reply is a summary of targets, succeeded and failed, plus one row per site with ok and either its data or its error. Some sites failing is a partial success. Only a run where every site failed comes back as an error.
06
Follow up on the failures
The agent reports which sites failed and why, then retries just those handles with an array. Tools that read one site, such as get_cache_status and get_health, do not fan out, so a fleet-wide check means one call per site.
Reference
| Fan-out tools | purge_cache, toggle_cache, update_settings, start_preloader, stop_preloader, optimize_site, run_benchmark |
|---|---|
| Selector | site: "all", or an array of handles |
| What all covers | Every site this connection can see in one workspace, minus revoked ones |
| Pacing | One after another, with a 1 second pause between live calls |
| Result | summary (targets, succeeded, failed) and a row per site: ok, then data or error |
| Partial failure | Reported per site; the call counts as an error only when every site failed |
| Workspace argument | Optional. Omit it for the default workspace; name one by its handle or id from list_workspaces |
| Reach | Fixed when you approve the connection; list_workspaces shows reachable: false for the rest |
| Not fan-out | get_health, get_cache_status and every other tool take one site per call |
| Slow tools | optimize_site takes up to 2 minutes per site; a target_score run is capped at 12 rounds per call |
| Read-only | The connection token with Read-only everywhere on, or a Viewer sign-in, refuses every fan-out write; run_benchmark works. An OAuth sign-in gets the scopes the client asks for |
| Hub daily pass | Re-verifies sites, snapshots cache status and hit ratio, reads stored reports, alerts by Slack webhook or email |
Rules worth keeping
- Confirm once, then call once. The Hub has no confirmation step of its own, so your client's approval prompt and a read-only connection (the connection token with Read-only everywhere on, or a Viewer sign-in) are the gates that matter. The Hub's "this will touch all N sites" is an instruction to the agent for purge, toggle, settings and preloader calls, and it does not name optimize_site.
- "all" is one workspace, never every workspace. To act in another workspace, name it. A write to a non-default workspace must name it, and an array of handles is looked up only in the workspace the call names.
- Read every result per site. Treat some sites failing as a partial result, and never accept a blanket "done" from the agent.
- Try a behaviour change on one site before you send it to the fleet. toggle_cache and update_settings on every site are one call, and each site checks values against its own settings, so a Pro setting can fail on a Free site.
- optimize_site on many sites is slow, because each site can take up to two minutes inside one call. Use dry_run first, or name a short list of handles.
- The Hub has no scheduler for agent prompts. Its own daily pass and alerts run without an agent, and an agent can only act when you send a prompt or your AI client runs one on a timer.
Good to know with Gemini CLI
Always allow on a tool or on the whole server, which the confirmation offers beside Proceed once, covers the site: "all" form of that tool too, and a trust setting on the server, or starting the CLI with --approval-mode=yolo, skips the confirmation prompt altogether. If you set trust on xspeedhub because the read calls were noisy, you also skip the prompt for site: "all" writes, and the only thing left between you and a fleet-wide change is the Hub's instruction to the agent, which is not a lock. Leave trust off for this server and do not use yolo for fleet work, or keep the connection read-only (the connection token with Read-only everywhere on) and widen it only when you need a fleet write. A read-only connection still lets you run run_benchmark across every site.
More prompts for this job
They work in any client connected to xSpeed Hub.
Use xSpeed Hub to list every site I can reach, and every workspace, and say which workspace you act in by default.
With xSpeed Hub, purge the cache on every site in my workspace and tell me which ones failed.
Use xSpeed Hub to run a benchmark on all my sites and rank them by cached response time.
Use xSpeed Hub to start the preloader on blog, shop and docs only.
With xSpeed Hub, turn page caching on for every site where it is off. Check each site first.
Use xSpeed Hub to stop the preloader everywhere. The servers are busy.
Ask xSpeed Hub to preview a safe optimize_site pass on every site with dry_run and summarize what it would change.
Use xSpeed Hub to purge every site in the client-acme workspace, not my default one.
Frequently asked questions
Keep going
Manage every site at once with other agents
More with Gemini CLI
Documentation
- Managing your fleet with xSpeed Hub
- Teams, roles and workspaces in xSpeed Hub
- Scoped AI connections in xSpeed Hub
- Connecting to xSpeed Hub
- How to write prompts for xSpeed Hub
- Gemini CLI + xSpeed
- Every AI agent that works with xSpeed
From the blog