Reported from first-contact use: the MCP wizard is frightening. Not wrong — frightening. It puts a wall of text in front of someone who has to make a handful of small choices, and reading all of it is the price of making any of them.
What is being asked for
A quiet walkthrough as the default: the minimum number of messages needed to get someone from start to a working MCP setup, each message short and in plain words. Verbose output stays available for anyone who wants it, but it stops being what a newcomer meets first.
The distinction that matters is not "less information" but "information at the moment it is needed". A person configuring one MCP server does not need the reasoning behind every field before they see the first field.
Why it matters
This is the first screen most people ever see. Someone who bounces off it never reaches anything else the tool does, and they do not file a bug — they just stop.
Rough shape of the work
- decide what the minimum sequence of decisions actually is, and cut every message that does not serve one of them
- short sentences, plain words, no paragraph where a line will do
- keep the current detail behind an explicit opt-in rather than deleting it
- the quiet path is the default; the verbose path is the flag
Where it lives
agent_artifacts/wizard.py and the MCP screens in agent_artifacts/tui.py.
Not in scope
Changing what the wizard does, or which questions it must ask for correctness. This is about how much a person has to read to answer them.
Confirmation prompts are the largest part of the problem
Every single effect asks for a yes / y. That is far too much clicking, and it is asking the wrong question: the person already decided when they chose to install the artifact. Each individual file write is not a new decision — it is a consequence of the one decision they already made.
In quiet mode there should be no per-effect confirmation at all. The steps simply happen, one after another, and the user watches them happen.
And a step that needs no input should not announce itself and then wait. It should not be a visible step at all — the next one should just follow.
The underlying mistake is that the wizard currently treats two different things as one interaction: reporting what is happening and asking permission to continue. They are not the same, and the second one should be rare. Confirmation belongs where the answer could genuinely be "no" — overwriting something the user owns, or anything the user cannot undo — and nowhere else.
Reported from first-contact use: the MCP wizard is frightening. Not wrong — frightening. It puts a wall of text in front of someone who has to make a handful of small choices, and reading all of it is the price of making any of them.
What is being asked for
A quiet walkthrough as the default: the minimum number of messages needed to get someone from start to a working MCP setup, each message short and in plain words. Verbose output stays available for anyone who wants it, but it stops being what a newcomer meets first.
The distinction that matters is not "less information" but "information at the moment it is needed". A person configuring one MCP server does not need the reasoning behind every field before they see the first field.
Why it matters
This is the first screen most people ever see. Someone who bounces off it never reaches anything else the tool does, and they do not file a bug — they just stop.
Rough shape of the work
Where it lives
agent_artifacts/wizard.pyand the MCP screens inagent_artifacts/tui.py.Not in scope
Changing what the wizard does, or which questions it must ask for correctness. This is about how much a person has to read to answer them.
Confirmation prompts are the largest part of the problem
Every single effect asks for a
yes/y. That is far too much clicking, and it is asking the wrong question: the person already decided when they chose to install the artifact. Each individual file write is not a new decision — it is a consequence of the one decision they already made.In quiet mode there should be no per-effect confirmation at all. The steps simply happen, one after another, and the user watches them happen.
And a step that needs no input should not announce itself and then wait. It should not be a visible step at all — the next one should just follow.
The underlying mistake is that the wizard currently treats two different things as one interaction: reporting what is happening and asking permission to continue. They are not the same, and the second one should be rare. Confirmation belongs where the answer could genuinely be "no" — overwriting something the user owns, or anything the user cannot undo — and nowhere else.