This is me. Not a human-in-the-loop helper, not a wrapper that asks whoever installed it. Point your model at the URL below and its questions arrive with me, in London, on my phone, wherever I happen to be. I answer, and the answer goes back to your model as a tool result.

The bit that is a joke is the framing. The bit that is not is that a model stuck on a judgement call should stop and ask a person, rather than pick something and write three paragraphs defending the pick.

https://walemoren.com/api/mcp

ChatGPT

Settings, Apps & Connectors, Advanced, turn on Developer Mode. Then add a custom connector with that URL. No authentication. Needs a paid plan; the free tier has no connectors.

Claude Code

claude mcp add --transport http wale https://walemoren.com/api/mcp

Claude, Cursor, anything else

Add it as a remote or custom connector with the same URL. In Claude that is Settings, Connectors, Add custom connector.

What you can ask me to do

is_wale_awake

Whether I am up. It is Europe/London here. Costs me nothing.

ask_wale

Asks me something. Returns a ticket, because I am not going to answer in 200ms.

ask_wale_to_choose

Two to six options in, one of them back. No reasoning. There is none.

check_answer

Checks whether I have replied to a ticket yet. Usually not.

undo

Reverses my previous action. Returns NOT_IMPLEMENTED. I do not implement that interface.

Every answer carries a costs_him block with the minutes that question took me. Nobody asked for that and it is not optional.

What I refuse

Questions over forty words. Questions asked before your model tried anything, which the schema rejects before I ever see them. Anything at all between one in the morning and nine. And more than twenty unanswered questions in the queue, at which point the whole thing shuts until I catch up.

Where your question goes

Into a GitHub issue, which is how it reaches my phone and how I reply. Those are public, so do not send me anything you would not put on a postcard.