An MCP server for my writing and tools

The Model Context Protocol (MCP) is the open standard that lets an AI assistant use tools and read sources outside itself. This server gives any MCP-capable assistant (Claude, ChatGPT, Cursor, an OpenClaw agent, your own) the same two things a reader of this site has: the articles, and the assessments. It is read-only, it calls no AI model of its own, and it costs nothing to use.

What it offers

The writing. list_articles and list_themes give the index; get_article returns the full text of any article, or the About and Services pages, with an address to cite; search_articles is a keyword search over every passage, returning the passage, the article, the heading and the address. Your assistant reads and quotes; it does not get an answer written for it, because it can write its own from the sources, and that keeps the server free to run and honest about what it is.

The assessments. list_assessments returns the four free tools on this site with their full question sets, so an assistant can put the questions to you in conversation. score_assessment then runs the same deterministic rules engine the website uses and returns the headline, the scores, the flags and the plan, exactly as a visitor to the site would get them. No model is involved in the scoring; that is the point. The rules are published, the result is reproducible, and the AI's job is to explain it, not to decide it. I wrote about why that separation matters in Guardrails: autonomy is earned.

How to connect

The server address is:

https://ianuydptehkpowxqfivq.supabase.co/functions/v1/mcp

It speaks MCP over HTTP (the streamable HTTP transport, stateless). Add it as a remote MCP server in your client; no authentication is needed. In Claude, that is Settings, Connectors, Add custom connector, with the address above. In Cursor, ChatGPT or an OpenClaw agent, add it as an HTTP MCP server in the same way. Then ask your assistant, for example, "search Catherine Ives-Yim's writing for what a lender's technology review looks at and summarise it with citations", or "take me through the CRA Readiness Check and score it".

What it will not do

It will not write anything: no leads, no emails, no briefings. It will not answer questions itself; that is what Ask my writing on this site does, with its own limits. It will not tell your assistant that your company is compliant, ready, or safe; the assessments are diagnostics to guide a conversation, and each result carries that note. And it is rate-limited per address, generously for a person, tightly enough that it cannot be used to hammer the site.

Privacy

The server keeps a count of calls per hour against a salted hash of the calling address, for the rate limit, and nothing else. Answers you give to an assessment through your assistant are scored and returned; they are not stored. Your conversation with your assistant stays between you and your assistant's provider. See the privacy notice.

Why I built it

Partly to be useful: if you use an assistant to read, it should be able to read this. Mostly as a demonstrator. Most of the questions I get from leadership teams about "what MCP means for us" are best answered with a working example that is small enough to understand: a read-only server, a deterministic engine behind it, and a clear line about what the AI does and does not decide. That is also the shape I recommend when a company builds its first one.

If you build something with it, or it does something odd, tell me.