May 6, 2026
The conference circuit is selling MCP for SEO as a verification tool. Stop pasting screenshots. Stop verifying numbers. Just ask. That pitch isn’t wrong. It’s also not the point.
Last Friday at Brighton SEO, I sat down in the live-stream booth with Gus Pelogia, Senior SEO Product Manager at Indeed, before his 11:20 talk on MCP servers. He’s been testing them across Ahrefs, Semrush, SE Ranking, and DataForSEO. After two years of building tools with APIs, and now MCPs, in my own stack, the lesson I wish I’d had at the start is the one his workflow makes obvious.
MCP for SEO is not a replacement for your judgment. It’s a permission slip to automate the recurring tasks that have been eating your strategic time.
If you do an SEO task more than twice a month, MCP is the case for automating it.
What MCP actually is, and what changed for SEO in 2025
MCP stands for Model Context Protocol. Anthropic announced it on November 25, 2024 as an open standard for connecting LLMs to data sources. Instead of pasting CSV exports into Claude or asking ChatGPT to "search the web", you give the model a server it can query directly. Anthropic shipped servers for GitHub, Postgres, Slack and Drive at launch.
The SEO ecosystem caught up fast. By April 2026, Ahrefs, Semrush, SE Ranking, and DataForSEO all run official MCP servers. Each one wraps the vendor's existing API and exposes it as tools the LLM can call. The pitch is identical across vendors: stop pasting screenshots, stop verifying every number by hand, just ask.
Gus is one of dozens of speakers running with that pitch this conference season. The pitch lands. The bigger story is what you do with the hours you get back.
How does MCP change a keyword research session?
Gus walked us through his actual workflow at Indeed.
His team is building landing pages for high-paying jobs by city and job family. The keyword universe for "high-paying jobs" is enormous, and most of it doesn't match the page pattern that Indeed serves. Trimming that list manually against the page set is a full afternoon every time the patterns expand.
With MCP, he asks the LLM to fetch the keyword set from Ahrefs and only return phrases that include a city name or a recognized job family.
Same data, same source, same numbers as the Ahrefs UI. The list comes back already filtered.
That sentence is the whole case. A task you used to schedule a Friday afternoon for is now a single prompt. If your team does that filtering for one new page set per month, you reclaim half a workday a month. If it's three page sets a month, you reclaim a full workday.
That's where the twice-a-month rule earns its keep. The cost of testing an MCP version of a recurring task drops below the cost of doing it manually faster than most teams realise.
What else can you chain through one prompt?
Gus chains his keyword filtering into a SERP-scanning step. The filtered keyword list becomes a batch of SERP queries against DataForSEO, scoped to the top ten US cities by population. DataForSEO returns the top ten search results per city per keyword, and the LLM summarises whether each SERP looks local or generic.
The old version of that work was: change Chrome's geolocation, copy a link, paste it, repeat. For one round of analysis. The MCP version is a single prompt that runs while you make coffee.
He also pulled an AI Overviews coverage report this way. He asked the LLM to walk through every keyword a domain ranks for and chart the percentage that shows an AI Overview, month by month, over the last twelve months. The output is a graphic he can drop into a stakeholder conversation about whether the AI Overview exposure is actually pulling traffic.
Three workflows surface from one conversation, and none of them are the marquee "ask the LLM about your SEO" use case the conference circuit keeps pitching. They're all the same shape: a recurring task with a known input and a known output, automated end-to-end.
Ready to start scaling your business?
Let's get started
Does MCP for SEO actually solve LLM hallucination?
Worth answering directly, because it's the question every SEO lead asks before they test anything.
Gus's test method was direct. For every example he ran, he opened the vendor's UI in another tab, queried the same keyword, and compared. He did this across all three vendors he tested, balancing his examples to avoid playing favorites. The numbers matched. Every time. So the answer is yes, when the LLM is querying through MCP and not improvising.
The reason is mechanical. The API is now in the loop, and the API doesn't have a creative writing mode. If you ask the LLM for a number that the underlying API can return, it returns the number. If you ask for a number the API can't return, the LLM has two options: tell you it can't, or guess. Prompt-side discipline is in noticing which path the model just took. That's the part of the pitch that holds.
What MCP for SEO won't show you
There is one gap worth knowing about before you build a workflow on top of it.
I asked Gus how he gets the paid view of a SERP through MCP. He laughed.
The reason is structural, not personal. DataForSEO's docs say the Regular SERP endpoint provides "paid and organic" results, but in practice (I've tested this on my own stack across two years of SERP API work) paid items show up rarely. SERP-scraping infrastructure sees a different SERP than a human browser does, and ads are particularly sensitive to that gap.
The other endpoint, the one built for ads, is advertiser-scoped. DataForSEO's serp/google/ads_search/* family pulls from Google's Ads Transparency Center. It tells you which ads a given advertiser has been running, with creative and timing.
It does not tell you which ads were running on a specific keyword in a specific location at a specific moment. So even if your MCP setup calls both endpoints, you cannot reconstruct the integrated SERP a human user actually sees.
This matters for one specific kind of question: "what does my customer actually see when they search this keyword in this market right now?" For that, you still want to look with your own eyes, in an incognito browser, from the right geolocation. MCP isn't replacing that yet.
For everything else (keyword research, organic positioning analysis, AI Overviews coverage, content gap analysis, internal linking audits, anything where a SERP-scrape with patchy ads is fine) MCP is the workflow.
The twice-a-month rule
Look at your last quarter of recurring SEO work. The keyword filtering you ran for ten landing pages. The SERP comparison you pulled for three priority keywords across five geos.
The AI Overviews coverage report you assembled for a stakeholder deck. The competitor backlink delta you eyeballed every fortnight. If any of those happened more than twice a month, that's an MCP or automation candidate.
If you do an SEO task more than twice a month, the cost of testing an MCP-driven version of it is lower than the cost of doing it manually for one more quarter. That's the rule. The conference pitch frames MCP as a hallucination fix. The real value is the hours back.
Hours back is the time you spend on work that LLMs can't do. The brainwork. The strategic sessions. Reading your client's product roadmap and figuring out which queries matter in two quarters, not last quarter.
Looking at a SERP with your own eyes and noticing what an API can't see. Talking to a customer success lead about why one segment is converting on a specific search intent that nothing in your dashboard explains.
Automating the recurring stuff is what makes time for the work where you actually earn your seat at the table.
What to do after reading this article
The integrated-SERP view is one of the things you'll keep doing manually for now. Most of the rest is fair game.
Listen
-
Hear the full conversation with Gus Pelogia on SEO Cast, recorded live at Brighton SEO 2026.
Head of SEO
As Head of SEO at Seeders, Hans helps companies scale smarter by combining search behaviour, paid performance, and conversion insights into systems that drive real growth. Recognised as a Top 20 Content Marketer by Small Business Trends, he brings a lean, experimental approach focused on building what actually works. Rather than relying on reports or one-off strategies, Hans focuses on creating marketing systems and tools that turn visibility into measurable business impact. His goal is simple: help clients move from scattered efforts to structured growth.
