When to Use (and Avoid) MCP Servers

Transcript

1. When to Use (and Avoid) MCP Servers Gus Pelogia SEO Product Manager SMX Advanced 2026
2. LLMs, 2024 Before MCP Keyword Search Volume (US) used cars 800k used cars near me 420k used cars for sale 125k
3. LLMs, 2026 Before MCP Keyword Search Volume Hallucinated Search Volume Real* used cars 800k 673k used cars near me 420k 450k used cars for sale 125k 301k *Matching SEO tool of choice
4. Model Context Protocol An open-source standard that enables AI models to securely connect with external tools In other words: APIs inside LLMs
5. Why MCPs are so powerful Gus Pelogia SEO Product Manager SMX Advanced 2026
6. Why MCPs are so powerful LLM MCP Find the top 5 "seo conference" related keyword by search volume Calls Sistrix/SE Ranking/Ahrefs keyword database to reply
7. Why MCPs are so powerful LLM MCP 1 MCP 2 Find the top 5 "seo conference" related keyword by search volume Calls Sistrix/SE Ranking/Ahrefs keyword database to reply Use DataForSEO and return the SERP results for US, UK and Ireland
8. SEO MCP Servers Timeline DataForSEO SE Ranking Ahrefs AlsoAsked May '25 Oct '25 Nov '25 Jun '26 Sistrix Profound Screaming Frog Sitebulb Aug '25 Oct '25 May '26 Sep '26
9. Keyword Research Ahrefs, Sistrix, SE Ranking, Semrush Ps 1 : Same tasks can be performed by different tools Crawling Screaming Frog, Sitebulb SERP Scraping DataForSEO, SE Ranking
10. Ps 2: All is happening inside Claude Code (Desktop)
11. 1 Extract PAAs from keywords and add to tracking prompts
12. Concept Run a keyword research, extract People Also Asked questions and place the PAAs into your prompt tracking tool
13. Check SE Ranking MCP and return the keywords [domain] ranks for in top 3 organic results Run the first ten terms by AlsoAsked MCP and return me the questions listed Now use SE Ranking MCP to add the first level questions into prompt tracking for AI Mode
14. Claude Code SE Ranking Regular UI
15. Claude Code Profound UI
16. Tool Shout Out: Sistrix
17. Sometimes, MCP is not the right tool When you need to find RAG dependency
18. Prompt webSearchQueries What foods do you eat in a keto diet? What is magnesium critrate used for? What does turmeric does for the body? Can I influence an answer?
19. Gemini Grounding API
20. You can still request your tracked prompts from another MCP Gemini has an MCP, but Claude created a script to run this webSearchQueries is the API field used. Not a probability score, only the result for this prompt, checked once Gemini Grounding API
21. 2 Use tracked keywords for live rankings on Google Maps
22. Concept Check your regular SERP keywords on GMaps, by city.
23. Ask which keywords / prompts are tracked. Select relevant ones Ask DataForSEO to scrape terms in list of locations Summarize most popular brands on [top 5] spots
25. I put 50USD back in February - still have $47 Tool Shout Out: DataForSEO
26. Sometimes, MCP is not the right tool When you need to scrape >20 SERPs
27. Live SERP Scraping at Scale
28. Use LLM to write a Pyhthon code to run as a skill or Google Colab It works, but it's slow. More than 20 SERPs, not worth waiting Claude will right a Python script to run it - and do a few at a time Live SERP Scraping at Scale
29. 3 Keyword patterns with multiple intents
30. Concept Take a broad keyword and organize in buckets, even when n-grams aren't useful
31. Pick a broad term that can have multiple intents, like "insurance" or "'restaurant" Slice into any relevant buckets: brands, intents, features Get a quick overview of demand instead of indivudual queries
33. 4 Gap analysis including rankings and existing pages
34. Concept Perform a content gap analysis using rankings and published URLs
35. Ask Claude to crawl [Domain A] and [Domain B] using Screaming Frog Layer over the [top 3] ranking keywords for each domain Find (1) pages I already have, but don't rank and (2) pages I don't have at all.
36. These are local Irish brands in my city - I'm not affiliated to them
37. 5 Audit and create structured data templates
38. Concept Identify Structured Data using Sitebulb's report, verify on Rich Results tool and generate new template
39. Look at Sitebulb's crawl for [domain] and audit my structured data results What SD types do I have? Which ones have mandatory vs optional fields missing? How could I improve or add new ones? Show me a few URLs inside Google rich results validation so I can check directly
41. 7 Use a crawl to summarize topics covered on a website
42. Concept Crawl using Screaming Frog directly from Claude to extract content insights
43. Use Screaming Frog to crawl [domain and find which names or topics are mentioned the most Create insights without manually looking at pages Cover topics, names, insights at scale
45. Sometimes, MCP is not required If you just need a quick crawl
46. Useful for some easy to read elements (e.g. <title> or status code) Doesn't replace SEO crawlers Can generate a script and crawl on Terminal or Colab Run a simple crawl
47. Run a simple crawl
49. 8 Compare similarity between AI Overview and organic results
50. Concept Run a crawl
51. Use DataForSEO to scrape AI Overview + SERP results. Saves as an .md file Store Gemini API key in .zshrc, so your key is not known by Claude Compare results
52. It works - but it's slow. AI Overviews + Five URLs = +13 min
53. AIO vs Organic AIO vs Sources Rank URL Similarity Rank URL Similarity 6 theguardian.com 0.8357 1 newsroom.heart.org 0.8286 4 newsroom.heart.org 0.8286 2 hopkinsmedicine.org 0.8271 2 hopkinsmedicine.org 0.8271 3 health.harvard.edu 0.7871 8 racgp.org.au 0.8268 4 healthline.com 0.7849 7 florre.co 0.8243 5 bhf.org.uk 0.7738
54. Is an MCP the right answer? Claude Colab Use MCP to get embeddings Run crawl on Screaming Frog with embeddings Create a skill to run on Claude Code Use Google Colab to get cosine similarity Next step: do I have pages with high similarity on my website?
55. guspelogia.com/smxadvanced