Sitebulb's MCP allows you to query crawl data in your AI assistant of choice - you can ask it questions and get it to build things for you. This guide is a primer on what the MCP is and how to get going with it. Once you start digging into it yourself, you'll find a whole host of things you can build.
What even is an MCP?
MCP stands for 'Model Context Protocol'. It's an open standard that lets an AI assistant talk directly to an outside tool, in this case Sitebulb. It works as a conduit: once Sitebulb is plugged in, the assistant can reach over and pull your audit data mid-conversation, rather than guessing or searching the web.
You don't need to understand any of that to use it. The practical upshot is the only bit that matters: connect it once, then ask questions in normal English and get answers from your real Sitebulb data.
What you'll need
A Sitebulb account with at least one project that's been crawled. The MCP reads finished crawls. It doesn't run them.
ChatGPT or Claude. It works the same way in both, but the setup steps for each are slightly different.
Connect the MCP
Connecting takes a couple of minutes and only happens once. We've written separate step-by-step guides for each client: Connecting to your AI assistant.
Follow whichever applies, then come back here for your first prompt.
Your first prompt
Start simple and build up. Try these in order:
See what's connected. "Using the Sitebulb MCP, show me my projects." You'll get a list of your projects and when each was last crawled.
Get the lay of the land. "Summarise my latest audit for [site]." Expect an overview: the site score, how many URLs were crawled, and the headline issues by severity.
Go deeper. "What are the biggest issues on that audit, and which would you look at first?" Now it's reading the actual hints and reasoning about them.
That's the whole loop. Ask, read, ask a sharper question. You're not learning a query language, you're having a conversation.
Make it visual
The quickest way to feel what the MCP gives you is to ask for a picture rather than a paragraph. Every report table comes back with a built-in steer on how it's best shown, as a table, a bar chart, a pie chart or a line chart, so when you ask to see something, the assistant already knows which fits.
Charting examples
"Chart the issues in my latest [project] audit by severity."
A count-by-category like this is flagged as a bar chart. If you ask for a proportion instead, the same data will render a pie chart.
"Show my latest [project] audit's hints as a pie by type."
And anything that moves across crawls, like the site score, comes through as a line chart.
"Plot how the site score for [project] has changed over the last few crawls."
One thing to remember: charts like these reflect a snapshot of the crawls you asked about, not a live feed. After the next crawl finishes, ask again to refresh it. When you want to go further, you can build out trends over time, wins-and-regressions boards, dashboards you keep coming back to - check out the 'What you can build' guide for inspiration.
How to talk to the MCP: worthwhile habits
A few small habits make the difference between a vague answer and a genuinely useful one.
Tell it to use the Sitebulb MCP. If you don't, the assistant may answer from its general knowledge or search the web instead of your data. Starting with "Using the Sitebulb MCP..." keeps it honest.
Name the project. Sitebulb MCP needs to know which project you mean to pull data from. You might have hundreds of projects, and potentially multiple for the same website. "Summarise the latest audit for the International Weekly project".
Be specific about dates and counts. "Compare the two most recent audits" or "give me the audit score from the crawl in June" gets you a tighter answer than an open-ended question.
Remember it reads finished crawls only. If you want fresh numbers, schedule or run the crawl in Sitebulb first, then ask. The MCP can't kick one off for you.
When in doubt: ask the assistant
If you aren't sure how to use it, ask the assistant what to ask. Tell it your situation - "Help me write a prompt for the Sitebulb MCP. I look after SEO for a mid-size ecommerce site, what could I ask you about my Sitebulb data?" - and it'll suggest prompts tailored to what might actually be valuable to you.
It's a quick way to find the useful questions without reading a manual.
When something's not right
Until you are familiar with the MCP and how it works, some of these things may come up as questions or issues:
"It says it can't find an audit." Check the crawl has finished, and that you've named the project the way it appears in Sitebulb
"I only got 50 URLs back." For any single issue, the MCP returns a sample of up to 50 example URLs, not the full list. That's deliberate, it keeps answers fast and readable. For every affected URL, use the export in the Sitebulb app.
"Can it run a crawl for me?" No. The MCP is read-only by design, so it can look but never touch. Schedule crawls in Sitebulb as usual, and the MCP will read them once they're done.
"It answered from the web, not my data." Remind it: "Use the Sitebulb MCP for this." Habit one, again.
"Is my data safe?" The MCP reads your Sitebulb data through your own authenticated connection, and only shares it with the AI client you've connected.
Next steps
Read through any or all of the following:
What the Sitebulb MCP can and can't do - explains what the Sitebulb MCP can read, and what it can and can't do.
Sitebulb MCP prompt library - Copy-paste prompts for the Sitebulb MCP, grouped by task.
What you can build with Sitebulb MCP - reports, dashboards, and combining Sitebulb with your other tools.