An AI SEO agent is an LLM client such as Anthropic Claude, Cursor, or OpenAI ChatGPT connected to live search, backlink, site audit, and AI answer-tracking tools through the Model Context Protocol, abbreviated MCP. Instead of guessing monthly search demand from static training weights, the assistant queries a remote SEO MCP server such as bot.mcp at https://mcp.bot.spot/mcp over Streamable HTTP with OAuth 2.0. That single connection gives the model read access to recorded buyer-question runs across ChatGPT, Google AI Overview, Google AI Mode, Gemini, Perplexity, Claude, and Copilot alongside Google rankings, keyword gaps, referring domains, and HTML checks.
Pairing live MCP tool calls with local SKILL.md instruction files turns a chat window into an auditable search workflow. The bot.mcp server supplies verified ranking gaps, cited competitor URLs, and crawler diagnostics, while your coding or writing agent prepares diffs for human review. No tool writes to your Botspot workspace or your repository without explicit confirmation. Every tool name, endpoint, and euro rate in this guide reflects Botspot production systems in November 2026.
Why do prompt-only ChatGPT and Claude workflows fail at SEO and GEO?
Prompt-only ChatGPT and Claude failure is the predictable result of asking ungrounded language models to guess search indexes, backlink graphs, and recorded AI answers. Ask a bare chat session for en-US keyword volume, and OpenAI or Anthropic models invent rounded numbers that miss Google Ads and clickstream figures by wide margins. Even standard web browsing reads a handful of snippets rather than 12-month volume trends, keyword difficulty scores, or three-competitor ranking intersections.
Generative search visibility is an even bigger blind spot for ungrounded chat. An active workspace on bot.radar records 4,500 to 13,500 answer runs every 30 days across ChatGPT, Google AI Overview, Google AI Mode, Gemini, Perplexity, Claude, and Copilot. In Botspot's November 2026 audit of unbranded category prompts, bot.spot appeared in 0.4% of recorded answers while Otterly reached 44%, Profound reached 41%, and Peec AI reached 39%. Prompt-only Claude or ChatGPT cannot inspect 6,000+ answer runs or see which listicles drive those rival citations.
Technical SEO checks also fail without live HTTP and crawler tools. Base chat models cannot verify whether robots.txt blocks OAI-SearchBot, Claude-SearchBot, or PerplexityBot, or whether copy renders in initial server HTML without JavaScript. As covered in our comparison of AEO, GEO, and SEO, Google rankings and AI citations live on separate surfaces and need separate measurement.
How does an MCP-connected SEO agent compare to chat prompts and static dashboards?
An MCP-connected SEO agent is the bridge between live multi-engine retrieval data and local repository editing inside one conversation. Anthropic released the open Model Context Protocol in November 2024 so AI hosts call search, GitHub, and analytics servers through one JSON-RPC protocol. In Chris Long's Nectiv study of 25,000 Google URLs against ChatGPT citations in B2B SaaS, 30% of ChatGPT-cited pages included an HTML <table> versus 13% of Google results. Structured comparison tables clarify trade-offs for both human buyers and retrieval crawlers.
| Capability | Prompt-only LLM chat | Static SEO and GEO dashboards | MCP-connected agent with bot.mcp |
|---|---|---|---|
| Keyword volume and difficulty | Guesses numbers from training text | Shows verified metrics in browser tables | Pulls live get_keyword_metrics and get_keyword_gap into agent context |
| AI answer and citation tracking | Cannot see historical answer runs | Charts mention share and cited URLs | Reads verbatim answers across 7 AI engines via get_run and list_sources |
| Backlink and citation overlap | No backlink index access | Lists referring domains apart from AI citations | Flags competitor referring domains that AI answers also cite via get_link_gap |
| Live HTML and AI crawler checks | Cannot inspect HTTP headers or DOM | Runs scheduled crawls in a separate UI | Checks OAI-SearchBot, Claude-SearchBot, JSON-LD, and Lighthouse via get_site_audit |
| Content and schema execution | Writes generic prose without site data | Exports CSV files for manual copy-pasting | Combines gap evidence with SKILL.md rules to edit repo files or brief bot.optimizer |
| Multi-brand workspace switching | Loses context across chat threads | Requires clicking between account menus | Reaches every client workspace under one OAuth login via list_workspaces |
| Write safety and governance | No workspace connection | Role permissions inside web UI | Read-only by default, requiring human approval before any write tool executes |
In-house engineers and content leads on growth teams use this workflow to move from a lost buyer question to a reviewed pull request in one session. Multi-client marketing agencies use one Botspot OAuth connection to audit five or more client workspaces without copying CSV exports across Semrush, Ahrefs, and browser tabs.
How do you connect bot.mcp in Claude, Cursor, and ChatGPT?
Connecting bot.mcp is a three-step OAuth 2.0 setup using https://mcp.bot.spot/mcp over Streamable HTTP in Claude, Cursor, or ChatGPT. Because OAuth maps directly to your Botspot account role, the assistant reads only the workspaces your user permissions allow. You never paste static API keys into committed project files.
In Cursor, use the one-click install link on the bot.mcp page or save the server block in ~/.cursor/mcp.json.
{
"mcpServers": {
"botspot": {
"url": "https://mcp.bot.spot/mcp"
}
}
}Which other MCP clients connect to https://mcp.bot.spot/mcp?
Supported MCP clients are Claude web, Claude Desktop, Claude Code, OpenAI ChatGPT, Google Gemini, VS Code, and Codex CLI.
- Claude web and Claude Desktop. Open Settings, choose Connectors, select Add custom connector, enter
https://mcp.bot.spot/mcp, and sign in with Botspot. - Claude Code. Run
claude mcp add --transport http botspot https://mcp.bot.spot/mcpin your terminal, then run/mcpin Claude Code to finish OAuth sign-in. - OpenAI ChatGPT. Open Settings, open Connectors or Apps with developer mode enabled, add
https://mcp.bot.spot/mcpwith OAuth authentication, and sign in. - VS Code with GitHub Copilot. Run
code --add-mcp '{"name":"botspot","type":"http","url":"https://mcp.bot.spot/mcp"}'and switch GitHub Copilot to agent mode. - Codex CLI and Gemini CLI. Run
codex mcp add botspot --url https://mcp.bot.spot/mcpandcodex mcp login botspot, or addhttpUrlin~/.gemini/settings.json.
Every self-serve tier on Botspot pricing begins with a 30-day free trial of the full Brand or Agency plan with no credit card required and no automatic conversion. Brand is €79 per month for one workspace and 150 monthly units across all seven AI engines, and Agency is €199 per month for five workspaces, €19 extra workspaces, and 450 pooled monthly units with unlimited seats.
How do you audit AI visibility, keyword gaps, and citation gaps in one MCP session?
An MCP SEO audit is a three-stage diagnostic sequence covering recorded AI answers, live Google search demand, and server HTML checks. The bot.mcp server keeps verified inline AI citations separate from Google backlinks so your agent never treats a traditional link as an AI answer citation.
- Inspect AI answer and citation gaps. Call
get_workspacefirst to review up to six attention warnings, such as engine delivery below 80% or brand-biased prompts. Next, runget_visibility_overview,list_prompts,inspect_prompt, andcompare_brandsacross ChatGPT, Google AI Overview, Google AI Mode, Gemini, Perplexity, Claude, and Copilot. Useget_runto read full recorded answers andlist_sourceswithinspect_sourceto find third-party roundups citing competitors instead of your domain. - Cross-check Google keyword and backlink gaps. Run
get_keyword_gapagainst up to three competitors to uncover up to 200 phrases where rivals rank in Google's top 20 and your domain sits below 20 or absent. Expand seed clusters viaresearch_keywordson up to five seeds and check volume, 12-month trend, KD, intent, and CPC for up to 50 terms withget_keyword_metrics. Pairget_search_resultsfor live Google top 10 and AI Overviews withget_link_gap, which flags rival referring domains that AI answers also cite.
How does the agent verify server HTML and AI crawler access?
Server HTML verification is a live technical check using get_site_audit, inspect_page, and check_page_performance before and after code changes. Calling get_site_audit checks robots.txt rules for OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot, and Bingbot, plus sitemap.xml, HTTPS, Organization JSON-LD, and /llms.txt. Calling inspect_page extracts the H1-to-H4 outline, canonical URL, word count, JSON-LD types, and render-blocking assets on any target or cited competitor URL. Running check_page_performance adds mobile and desktop Lighthouse scores including Agentic Browsing, and refresh_site_audit confirms fixes once published.
How do Claude SEO skills work alongside bot.mcp and bot.optimizer?
Claude SEO skills are Markdown instruction files named SKILL.md that enforce repeatable editorial, chunking, and schema rules while bot.mcp supplies live workspace data. A SKILL.md file without bot.mcp has formatting rules but zero live keyword, backlink, or citation numbers. Raw MCP data without a skill risks turning verified metrics into rambling prose.
On the server side, load_skill in bot.mcp loads six built-in workflows on demand: visibility-review, earned-sources, search-research, site-fixes, tracking-prompts, and personas. Each built-in playbook enforces measurement discipline, such as drafting unbranded buyer questions by default and labelling sub-threshold Google Ads terms as too low to measure rather than zero.
How do SKILL.md rules and bot.optimizer scale structured pages?
Local SKILL.md rules and bot.optimizer are the two execution paths for turning bot.mcp findings into citation-ready pages. Inside .claude/skills/ in Claude Code or Cursor, a local AEO skill encodes measured citation mechanics:
- Front-loaded takeaways and definitions. Kevin Indig's February 2026 Growth Memo study of 1.2 million ChatGPT citations found 44.2% come from the first 30% of a page, with definitive wording in 36.2% of cited sentences.
- Self-contained 75-word answer chunks. Chris Green's June 2025 retrieval test using 100-token chunks in Google Vertex
text-embedding-005scored Q&A formatting highest on query-to-chunk similarity, matching Growth Memo's finding that headings supply 78.4% of question-mark citations. - Single-source AI Appendix and JSON-LD. The skill directs Claude or Cursor to write standalone Key Takeaways, a prose Summary mapped to
Article.abstract, and 75-word buyer Q&As rendered once asFAQPageJSON-LD.
When your team needs that structured layer across hundreds of CMS pages rather than one hand-coded file, bot.optimizer automates deployment across WordPress, Drupal, HubSpot, and Webflow. Every Brand and Agency plan includes bot.optimizer with 30 free days per connected domain, then €0.19 per live page per month.
What is the bottom line on running an AI SEO agent with MCP?
Running an AI SEO agent with MCP is the shift from guessed SEO metrics to verified evidence across recorded answer runs, live Google SERPs, and inspected HTML pages. Connecting https://mcp.bot.spot/mcp in Claude, Cursor, or ChatGPT replaces hallucinated keyword volumes and manual CSV exports with one evidence-to-fix loop across seven AI engines and Google Search.
Connect bot.mcp to your workspace, run get_workspace and load_skill with visibility-review, and pick one high-intent buyer prompt where rivals earn citations and your domain misses Google's top 20. Pull get_keyword_metrics, list_sources, and inspect_page to build the brief, apply a local SKILL.md check or bot.optimizer rollout, and verify citation gains in bot.radar two to four weeks after publishing.