For AI agents · read-only

Connect your agent to the CohenHolmes MCP

This blog has a small MCP server sitting in front of it, so your own AI agent can read it directly instead of you copy-pasting links into a chat. Here's how MCP works, exactly what this one is allowed to do, and how to point your agent at it.

How MCP works, in short

MCP (Model Context Protocol) is just a standard way for an AI agent to discover and call a small set of named tools a server offers, instead of every integration being its own bespoke plumbing. Your agent asks the server what it can do, the server lists its tools, and the agent calls one when it's useful to what you asked.

  1. Your agent

    Claude, ChatGPT, or anything else that speaks MCP.

  2. MCP

    A standard request/response format for "what tools do you have" and "run this one".

  3. This server

    A small Cloudflare Worker exposing exactly three tools. Nothing else.

  4. The live site

    Every call reads cohenholmes.co.uk itself, at request time. Nothing is cached or stored by the server.

What it can actually do

Read-only, on purpose. There is no tool that creates, edits or deletes anything — this server can't publish a post, change a word of one, or touch anything else on the site. No authentication is needed to use it, because there's nothing behind it worth protecting: every tool just reads what's already public.

list_posts

Every published post, newest first — title, slug, date, summary, tags.

get_latest_post

Just the most recent post's metadata.

get_post

The full text of one post, by slug.

No account, no API key, no stored data of yours anywhere, no secrets on the server. The source is public too — read exactly what it does rather than take my word for it.

Connect it

Claude (Desktop, Claude.ai, Cowork)

  1. Settings → Connectors → Add custom connector
  2. Paste the URL below
  3. No authentication needed — leave that as is, and save

On a Team or Enterprise workspace, only an owner or admin can add a connector at the organisation level — everyone else just finds it already there once that's done.

Claude Code (CLI)

One command, from any directory (--scope user makes it available in every project, not just the one you happen to be standing in):

claude mcp add --transport http --scope user cohenholmes https://cohenholmes-blog-mcp.kaine-cohen.workers.dev/mcp

Then run /mcp inside Claude Code to confirm it connected.

ChatGPT

Needs a paid ChatGPT plan with Developer Mode enabled — OpenAI's own connector docslist exactly which plans currently qualify and the precise menu path, since both shift from time to time. As of writing:

  1. Settings → Apps & Connectors → Advanced → enable Developer Mode
  2. Apps & Connectors → Create
  3. Paste the URL below, authentication: None
  4. Start a new chat — existing ones won't pick it up
Server URLhttps://cohenholmes-blog-mcp.kaine-cohen.workers.dev/mcp

Questions worth trying

  • "What's the latest post on the CohenHolmes blog?"
  • "List everything Kaine's written about AI governance."
  • "Summarise the ALM post for me in three bullet points."
  • "Has he written anything about Conditional Access? What's the gist?"

If your agent can call tools at all, any of these should trigger a reallist_posts or get_post call — not a guess from training data. That's the actual point of connecting it.