Blog
Notes on MCP, on Claude and ChatGPT as an analytics frontend, and on building mcp-analytics solo. Read in German.
PerplexityBot: Verification, robots.txt, and the Controversy
What PerplexityBot is, its user agent and IP ranges, the stealth-crawling reports, whether to block it, and how to see its hits on your own site.
OAI-SearchBot: The ChatGPT Search Crawler You Shouldn't Block
What OAI-SearchBot is, how it differs from GPTBot, why blocking it removes you from ChatGPT search citations, and how to watch its activity on your site.
GPTBot Explained: OpenAI's Crawler, robots.txt, and Blocking
What GPTBot is, the exact user agent and IP ranges, what blocking it in robots.txt opts you out of, and how it differs from OAI-SearchBot and ChatGPT-User.
ClaudeBot: What Anthropic's Crawler Does (and How to Verify It)
What ClaudeBot is, its exact user agent, how to verify real ClaudeBot traffic, whether to block it in robots.txt, and how to see its hits on your site.
CCBot: What Common Crawl's Bot Does and Who Trains on Its Data
What CCBot is, Common Crawl's monthly crawl pattern, why its corpus feeds many AI training sets, whether to block it, and how to see its hits on your site.
Bytespider: ByteDance's AI Crawler and Its robots.txt Problem
What Bytespider is, ByteDance's crawler behavior, the documented robots.txt compliance reports, how to actually block it, and how to spot it in your traffic.
MCP vs RAG: Different Problems, One Stack
MCP vs RAG: RAG gets knowledge into context, MCP gives models live tools. Where each fits, where they compose, and a decision table for your AI stack.
MCP vs API: What's Actually Different (With a Real Example)
MCP vs API is the wrong dichotomy: MCP is the layer that makes your API usable by LLMs. The real differences, a comparison table, and a production example.
MCP Inspector: A Practical Guide to Debugging Your MCP Server
How to use MCP Inspector in 2026: install, connect to local and remote servers, exercise tools, debug OAuth, and decode the most common errors.
LLM SEO: Get Cited by ChatGPT and Perplexity - and Prove It
LLM SEO in 2026: what actually influences ChatGPT and Perplexity citations, what is folklore, and how to measure whether your work does anything.
How to Track ChatGPT Traffic to Your Website (2026)
Three layers to track ChatGPT traffic: referrers and UTMs, AI-mediated visits that carry no referrer, and crawler activity that predicts citations.
Best MCP Servers in 2026: A Curated, Opinionated List
The best MCP servers worth installing in 2026, by category: dev tools, docs, browser automation, data and analytics. Curated, with honest disclosure.
AI Traffic Analytics: Measure Crawlers, Agents, and AI Visitors
AI traffic analytics explained: why GA4 hides bot and agent visits, an 8-class taxonomy for crawlers and AI-mediated traffic, and how to monitor it.
MCP OAuth: Every Bug We Hit Building a Remote MCP Server for Claude and ChatGPT
Three weeks of building OAuth 2.1 for a remote MCP server, most of it spent on client quirks. The 302 vs 303 bug. Lowercase bearer. Origin: null. CSP form-action. The /mcp suffix discovery quirk. Every quirk that made our connector silently die.
llms.txt Explained: The Robots.txt for AI Search (with Real Examples)
What llms.txt is, why it exists, what it isn't (it's not robots.txt for AI), how to write one for your site, and what AI clients actually use it for in 2026.
Claude MCP Setup: A Practical Guide for 2026
How to wire up MCP servers in Claude Desktop, ChatGPT, and Cursor. With the OAuth and CORS bugs that actually trip people up, written by someone who shipped a remote MCP server to production.
Web analytics with no dashboard. Ask Claude.
Enter your email, click the verify link, paste the tracking snippet, ask your stats in Claude or ChatGPT. Free up to 100k hits/month, no credit card.
Create account →