Every website has a front door for people.
Soon every person will have one for agents.

{
  "mcpServers": {
    "max-postnikov": {
      "type": "http",
      "url": "https://postnikov.ai/mcp"
    }
  }
}

codex mcp add max-postnikov --url https://postnikov.ai/mcp

Build me a context pack: six markdown files that describe me well
enough for a stranger's AI agent to answer questions about me
accurately, without inventing anything.

SOURCES — public only, in this order:
1. My website, every page. Read its JSON-LD if there is any: it is
   the most reliable source of names, roles and links.
2. My public writing: blog, newsletter, long-form posts.
3. My public talks, podcasts and conference bios.
4. Anything else I clearly published myself.

Do not use anything I did not publish. If a fact is not in a public
source, leave it out — a gap is cheap, an invented fact is not.

PRODUCE exactly these files, in English, markdown, no front matter:

  who.md       Who I am: role, city, the arc of my career, what I
               actually work on, what I am not. Include the public
               failures — they carry more signal than the wins.
  now.md       What I am doing right now: current projects, current
               focus, open experiments. This file goes stale first
               and is meant to be rewritten often.
  speaking.md  Talk topics with lengths and formats, stage history,
               podcasts, the languages I present in.
  services.md  The ways someone can work with me and what they get
               out of it. No prices, ever — write "pricing on
               request" instead.
  channels.md  Every public channel and link, plus one line on which
               channel suits which kind of request.
  writing.md   My key articles and essays: title, link, and one line
               on what each one argues.

RULES:
- Write for an LLM reader, not for a recruiter. Plain declarative
  sentences, concrete facts, dates, numbers, links. Drop every
  marketing adjective: "experienced", "passionate", "results-driven"
  are noise and cost me credibility with a careful reader.
- Keep each file under about 100 lines. This is a reference, not
  a memoir.
- Nothing about family, health, money, home address, or anything
  I would not put on a billboard. In doubt, leave it out.
- Mark anything you inferred rather than read, inline, like this:
  [? VERIFY — inferred from <source>]. I will check those by hand.
- End each file with a "Last updated: YYYY-MM-DD" line. Ask me for
  today's date if you are not certain of it.

BEFORE YOU WRITE, show me two lists: the sources you found, and the
things you looked for and could not find. I would rather fill a gap
myself than have you paper over it.

Here is my context pack: <attach the files>. This is what a
stranger's AI agent will read to answer questions about me. Audit it
hard. I want the problems, not encouragement.

Check five things, in this order:

1. PRIVACY. Flag anything that should not be public: family, health,
   money, addresses, unshipped plans, other people named without an
   obvious public reason, prices, deal sizes. Quote the exact line.

2. TRUTH. For every concrete claim — dates, numbers, titles, roles,
   links — try to confirm it against a public source. Give me two
   lists: claims you could not confirm, and claims that contradict
   what you found. Do not fix them; I will.

3. STALENESS. Which lines describe a state that has probably moved
   on? Anything with "currently", "this year", "recently"; any
   project that may have shipped or died since. Rank them by how
   wrong each would look if read today.

4. GAPS. A stranger's agent needs to answer five questions: who is
   this person, what are they doing now, what are they good at, are
   they a fit for a specific ask, and how do I reach them. Which of
   those cannot be answered from these files alone? Name the missing
   fact, not the missing file.

5. NOISE. Marketing adjectives, hedges, things repeated across
   files, sentences an LLM would have to guess at. Quote the ten
   worst lines and rewrite each as one plain sentence.

Then a verdict, in two sentences: reading only this, could you answer
a stranger's question about me without inventing anything? If not,
name the single change that would fix it.