A failure that crashes is cheap. A failure that reports success is expensive: the screenshot looks fine, the command exits zero, the endpoint returns 200, and the bug ships.

This server exposes verifyfirst.dev — a registry of 48 such failures, organised by the instrument that missed them rather than by the technology involved, because the instrument is what you know before you know the bug.

Install

uvx verifyfirst-mcp          # run it directly
pip install verifyfirst-mcp  # or install it

claude mcp add verifyfirst -- uvx verifyfirst-mcp

Or any MCP client:

{
  "mcpServers": {
    "verifyfirst": { "command": "uvx", "args": ["verifyfirst-mcp"] }
  }
}

Tools

Tool	What it does
`from_symptom(description)`	Start here when something is wrong but you do not know why. Describe what you see — "deploy ran but nothing changed", "200 but the data is wrong" — and get the failures that produce that appearance, each with its check.
`blind_spots(instrument)`	The one to reach for mid-task. What an instrument cannot see, terse. Takes loose names — `curl`, `200`, `pgrep`, `systemctl`, `stdout`, `playwright` all resolve.
`list_instruments()`	All six instruments.
`get_instrument(id)`	Full detail plus every recorded failure it missed, each with its check.
`search(query)`	Substring search across the registry.
`get_entry(id)`	One entry, e.g. `NS-001`.
`get_protocol()`	A five-step checklist for claiming work is done.

Example

> blind_spots("screenshot")

screenshot (A rendered image) is blind to:
  - Time. A still frame cannot distinguish 'renders one frame then stops'
    from 'renders one frame correctly'.
  - Whether scripts ran at all, as opposed to running and producing this.
  ...

Notes

Zero dependencies — Python standard library only. Works offline from a bundled copy of the registry; pass --remote to read the live one, which falls back to the bundle on any failure.

Every entry in the registry is a failure that genuinely occurs, marked observed (diagnosed first-hand) or documented (primary source cited and the check reproduced). None are hypothetical.

CC0-1.0 — public domain. Maintained by Zion Labs.