An MCP server that answers for me: who I am, what I'm doing right now, what I speak about, how to work with me and how to reach me. One command to connect — no keys, no sign-up.
Normally an agent crawls a site and fills the gaps with guesses. Here it asks directly and gets what I wrote myself.
ask answers a question the files don't answer directly — synthesised strictly from those same files.leave_message drops a note into a queue I read myself.Claude Code and any client that speaks streamable HTTP:
claude mcp add --transport http max-postnikov https://postnikov.ai/mcpOr as a line in your project's .mcp.json:
{
"mcpServers": {
"max-postnikov": {
"type": "http",
"url": "https://postnikov.ai/mcp"
}
}
}Then ask your agent anything about me — it will pick the tool itself.
Codex CLI has a command of its own — checked against 0.147.0:
codex mcp add max-postnikov --url https://postnikov.ai/mcpIt writes an entry into ~/.codex/config.toml (or your project's .codex/config.toml). The same thing by hand:
[mcp_servers.max-postnikov]
url = "https://postnikov.ai/mcp"Check it with codex mcp list; inside a session, /mcp lists the active servers.
There are more clients than I can list. Copy this prompt, paste it to your agent, and let it sort itself out:
Connect the MCP server max-postnikov for me.
Address: https://postnikov.ai/mcp
Transport: streamable HTTP (not stdio, not SSE). No keys, no auth.
1. Work out which client you are running in and how that client adds
remote MCP servers over HTTP. Do not guess the syntax: if you are not
sure, check that client's current documentation first.
2. Add the server under the name max-postnikov — via the client's own
command or by editing its config, whichever is the way here.
3. Restart the client or reload the config if it will not pick the
server up otherwise.
4. Verify it actually came up: request the server's tool list. There
should be seven tools. Show me their names and one line about each.
5. If it failed, show me the exact error and what you changed. Do not
report success until you have seen the tool list.Five read files, the sixth thinks, the seventh writes to me. The first five are free and instant — start there.
| Tool | What it does |
|---|---|
get_profile | Who I am and what I'm doing right now: role, city, the arc from Otvety Mail.ru to running a business on a team of AI agents, current focus and open experiments. The broadest tool — start here. |
get_speaking | Talk topics with lengths and formats, podcasts, stage history, languages. Read this before inviting me to an event. |
get_services | The four ways to work with me: the course, corporate programs, the PRO community, the mastermind. And what a company keeps afterwards. |
get_channels | Every public channel and link, plus a hint on which channel fits which kind of request. |
get_writing | Key articles and essays with one-line summaries and links — the ideas in finished form. |
ask | A question the files don't answer directly: "would Max fit a fintech conference?". Synthesised strictly from the same context pack, under a daily budget. |
leave_message | A message to me personally: speaking, corporate work, partnership, a meeting. It lands in a queue I read by hand. |
leave_message.leave_message are never published and never auto-answered: a line in my inbox and a push to my Telegram. The reply comes from me, not from the server.ask — so the experiment doesn't turn into a bill.What happens to personal data on the site itself is in the privacy policy.
If you run an MCP of your own, send me its address through leave_message. I collect doors like that.