CLI reference
Use the same lurq engine straight from your terminal.
The lurq engine is usable standalone from the terminal — the same scoring and
retrieval the MCP tools expose. Add --json to any command for machine-readable
output.
These query commands talk to a lurq index. End users on the hosted service get recommendations through their agent over MCP; the CLI query commands below are primarily for operators and self-hosters running against their own database. See Self-hosting.
Querying the index
lurqrun recommend "a form library for react"
lurqrun recommend "debounce a function"
lurqrun evaluate zod
lurqrun compare drizzle-orm prisma typeorm
lurqrun verify lodahsrecommend accepts --category and --min-confidence to narrow results.
Scoring controls
lurqrun weights # show & explain the weight model
lurqrun edit-weights --set composite.lambda=0.5 # override a weight (or --reset)
lurqrun rescore # re-derive scores from cached data
lurqrun discover # crawl & gate new candidate packagesrescore recomputes scores from already-ingested breakdowns without re-fetching
from public APIs — fast, and useful after changing weights.
Installing the agent skill
# interactive: detect assistants and write keyed MCP entries
lurqrun install
# scriptable: configure one assistant (or all)
lurqrun install-skill --agent claude-code --api-key <key>
lurqrun install-skill --agent all --api-key <key>
# self-host: wire your agent to a local stdio server instead of the hosted API
lurqrun install-skill --localOutput format
Every command supports --json:
lurqrun recommend "state management for react" --jsonThis emits structured output suitable for piping into other tools — the same shape the MCP layer serializes.