The Claude Fast Blog Is Now an MCP Server
Your AI can now read all 180 Claude Fast posts over MCP. Included with every kit: one key, one command, the whole corpus.
Agentic Orchestration Kit for Claude Code.
Every post on this blog exists to change something in your setup: your CLAUDE.md, your agent definitions, your hooks, your model routing. And until today, the only way to get that guidance into the agent doing the work was you, manually, one paste at a time.
That is now a connection instead of a chore. https://claudefa.st/api/mcp is live. It is a remote MCP server that serves the entire Claude Fast blog as agent-readable knowledge, and it is included with every Claude Fast kit.
What Changed
Your AI reads the blog directly. Not a summary of it, not a cached copy of one post you remembered to paste, but the live corpus: 180 posts across 53 knowledge topics, fetched on demand, mid-session, while it is working on your actual code.
Four read-only tools do it. list_topics shows what the corpus covers. list_posts returns dense summaries so the model can pick without paying to read. get_post fetches one article as clean markdown with a metadata header and a Related posts footer. search_posts handles the exact phrase or error string that a taxonomy would never surface.
The practical shape of that, in a real session:
The agent searches, reads two posts, and hands you a diff of your own file. You never opened a browser tab.
Why an Index, Not a Sitemap
The interesting part is not the endpoint. Wrapping a blog in HTTP is trivial. The interesting part is what the endpoint retrieves against.
A sidebar tree tells you where a post is filed. That is useless to a model, which does not care about our information architecture and cares enormously about which three of 180 posts address the thing it is stuck on. So the MCP does not serve the navigation tree. It serves a purpose-built knowledge index: 53 logical topics with many-to-many membership, a dense per-post summary written specifically for a model deciding whether the full article is worth its context, key concepts, and related-post links.
That is why the survey step is cheap and the fetch step is deliberate. Your agent spends a small amount of context learning what exists, then spends real context only on the posts that earn it. Context engineering, applied to documentation retrieval.
An integrity suite keeps it honest: a post is not published for the MCP until it has an index entry, and our tests fail if we forget one.
Nobody Else Has This
We looked. Plenty of vendors ship an MCP server for their product. We do not know of another Claude Code blog or knowledge site that exposes its own writing this way.
That gap says something about how documentation is still built. Everyone writes for a human with a browser, then treats agent consumption as scraping to be tolerated. The reader has changed. A meaningful share of the people acting on this blog are not people, they are agents working on someone's repository, and they deserve a real retrieval surface instead of HTML to guess at.
We would rather be early on that than tidy.
Connect It
Three steps, about a minute.
- Sign in at claudefa.st/account with your purchase email (any active kit unlocks the MCP). Magic link or Google.
- Generate your API key in the "Your API Key" card. It renders the full connect command with your key already substituted.
- Run it:
Replace YOUR_API_KEY with your actual key, exactly as copied. No angle brackets around it, or the server sees the brackets as part of the key and returns a 401.
Claude Code, Claude.ai, and Cursor all connect to the same URL. The full walkthrough, the four tools in detail, and the troubleshooting list live in Connect Your AI to the Claude Fast Blog.
What We Are Watching
The surface is deliberately small: four read-only tools over published content, no rate limits, one key per account that you can rotate whenever you want. We want to see how people actually query a knowledge base before deciding what it should grow into.
If you are new to MCP in general, MCP fundamentals covers how servers get discovered and invoked, and the servers worth installing is where to go next. If you want to build one, custom integrations has the server side.
Go generate a key. Then ask your agent to fix something using our guidance and watch it fetch what it needs on its own.
Last updated on