# How looot works, in depth

Discover the exact operation, inspect it, run it, and get an honest record after -- the same four steps every capability in the catalog goes through.

## Discover, inspect, run, audit

discover finds candidate operations for what you're trying to do. inspect reads one operation's real contract before anything runs. run executes it: quoted, reserved, and settled once. The record after is an honest account of what happened, not a claim.

## What a run passes through

Everything a run passes through is visible: the quote, the reservation, the provider call, the settlement, and the evidence written to a hash-chained audit log. Nothing you can't see.

## One MCP connection, however large the catalog gets

The tool surface stays fixed at six tools no matter how many providers or operations the catalog adds. No SDK to install -- call the tool, get the run.

## For agents

The full machine-readable tool contract lives at https://api.looot.ai/llms-full.txt; this block is a shorter pointer, not a second copy.

- **run lifecycle**: discover -> inspect -> run (quote, reserve, settle) -> runs_get (evidence)
- **tool surface**: fixed at six tools: discover, inspect, run, runs_get, runs_list, balance