WordPress and MCP: How AI Assistants Read Your Site Stats
Ask your assistant which site lost traffic last week and get an answer from your real numbers. That is what an MCP server does, and it takes one URL and a key.
On this page
WordPress MCP, in the sense most site owners mean it, is an AI assistant answering questions from your site’s real data: “which of my sites lost traffic last week”, “what are the top pages on the shop this month”, “is anything down”. The Model Context Protocol is the standard that makes that possible. A server exposes your data as tools; the assistant calls them when you ask. This post explains the pieces without jargon, using the DevDome MCP server as the example because we make it.
What MCP actually is
An AI assistant on its own knows nothing about your site. MCP gives it a way to find out. An MCP server publishes a list of tools, each with a name, a description and the arguments it takes. When you ask a question, the assistant reads the list, picks the tool that fits, calls it with the right arguments, and writes its answer from the result.
For analytics, the tools are the reports. The DevDome server exposes ten of them, each mapped one-to-one to a dashboard report: headline stats, time series, pages, referrers, countries, devices, visitors, bots, site health and so on. Ask about traffic, the assistant calls the stats tool. Ask about bots, it calls the bots tool. You never see the call; you see the answer.
How a connection works
One URL and your API key. The DevDome server is a stateless proxy over your account at a single address; there is no install, no OAuth flow and no session to manage. The client sends the key as a bearer token in a header, or, for clients that cannot send headers, as a parameter on the URL.
Setup differs by client:
- Claude Code adds it with one command naming the URL and the header.
- Claude.ai and ChatGPT add it as a custom connector in settings; since their dialogs cannot send headers, the key rides in the URL.
- Cursor, Windsurf, VS Code and Zed take a small JSON block naming the URL and header.
- Anything else can prove the connection with a single curl request listing the tools.
The full per-client instructions are on the MCP docs page.
What to keep in mind on keys
The key is the credential. Whatever it can read, the assistant can read, and anyone who gets the URL with the key in it can read too. Three habits:
- One key per assistant. Create a dedicated key for the connection rather than reusing the one in your scripts.
- Treat the keyed URL as a secret. Do not paste it into shared documents or screenshots.
- Revoke when done. Stop using the assistant, revoke the key. Making a new one takes seconds.
The key reads reports. It cannot change tracking, delete sites or alter settings, so the worst case is disclosure of your stats, not damage to the site.
What the assistant does with it
The useful questions are the ones that would take several dashboard visits:
- “Compare human pageviews across all my sites for the last 30 days and flag any that dropped more than 20 percent.”
- “Which referrers sent clicks to affiliate links on the shop last week?”
- “Is any site reporting downtime or a certificate expiring soon?”
- “Which bots hit the blog most this month and why were they classified as bots?”
Each is a few tool calls the assistant makes for you. Rate limits apply, 60 requests per minute per key, so an assistant that fans out many calls at once may pause briefly; it resets within a minute.
MCP in the cloud versus abilities in WordPress
Two related things are worth keeping apart. The MCP server above runs over your DevDome account in the cloud; any client can use it without touching the site. Separately, WordPress 6.9 introduced an Abilities API, a way for plugins to register actions, with permissions, that AI tools can call inside WordPress. Several DevDome plugins register abilities there: Bot Protection, Admin Cleaner, Country Blocker and Activity Log among them, so an assistant working inside your WordPress install can query logs or adjust settings through the plugin’s own permission checks.
Cloud MCP reads your reports from anywhere. In-WordPress abilities act on the site from inside. Both are the same idea, tools an assistant can call, at two different layers.
When the connection does not work
Most failures are one of three things. The client cannot connect: the server answers plain streamable HTTP with JSON-RPC and no OAuth or session, so check that the client is using the HTTP transport rather than expecting a local command, and prove the server with the curl example on the docs page. The assistant says it has no tools: the key is missing or wrong in the header or URL, so recreate it and paste it again. The assistant gets rate limited mid-answer: it fanned out many calls at once and hit the 60-per-minute limit, which resets within a minute; ask again or narrow the question. If the curl call lists tools and the client still fails, the problem is the client’s configuration, not the server.
Try it in five minutes
Create a key, add the URL to your assistant, and ask it to list your sites. If it answers with your real site names, the connection works. Then ask the question you usually answer by opening three dashboards, and see whether you still need to.
Disclosure: DevDome publishes this blog and makes the products it mentions. We describe what they do rather than promise results.
Key takeaways
- MCP is a standard way for an AI assistant to call tools. An MCP server exposes your data as tools the assistant can use when you ask a question.
- For analytics, each tool maps to a report: stats, pages, referrers, countries, bots, health. The assistant picks the tool, you read the answer.
- Connection is one URL plus your API key. The key is the credential; treat the URL that carries it as a secret.
- The assistant sees what the key can see. Use a key per assistant and revoke it when you stop.
- WordPress 6.9 adds an Abilities API that lets plugins expose actions to AI tools inside WordPress itself.
Sources
- Model Context Protocol specification : the standard and its transports
- WordPress Developer Resources: Abilities API : exposing plugin actions to AI tools inside WordPress
- Anthropic: Connect Claude to tools with MCP : adding a custom connector in Claude
Links last checked October 17, 2026.
Frequently asked questions
What is MCP in plain words?
The Model Context Protocol is a standard that lets an AI assistant discover and call tools provided by a server. Instead of pasting numbers into a chat, you connect the assistant to a server that answers questions from your real data when the assistant asks.
Which assistants support it?
Claude in its desktop and web apps and Claude Code, ChatGPT through developer-mode connectors, Cursor, Windsurf, VS Code, Zed and other MCP clients. Setup differs per client; some send the key as a header, some can only carry it in the URL.
Is it safe to give an assistant my analytics key?
The assistant can read whatever the key can read. Use a dedicated key for the connection, do not share the URL that carries it, and revoke the key when you stop using the assistant. The key cannot change your tracking; it reads reports.
Does it work on my WordPress site directly?
The DevDome MCP server runs in the cloud over your account, so it works from any client without touching the site. Separately, WordPress 6.9's Abilities API lets plugins register actions that AI tools can call inside WordPress; several DevDome plugins register abilities there.