MCP tools

Once lurq is installed, your agent can call these tools over MCP. Every response is compact (to stay cheap in tokens) and carries a dataAsOf timestamp so the agent knows how fresh the underlying evidence is.

recommend

Best current packages for a described need.

• Input: a natural-language description of what you need.
• Output: up to 5 packages, each scored, with a confidence label.
• Filters: category and min-confidence.

"a form library for react" → react-hook-form, @tanstack/react-form, …

Use this when the agent knows what it needs but not which package to reach for.

evaluate

A full evidence read for a single package.

• Input: one package name.
• Output: the complete score breakdown, advisories, and a usage guide.

Use this when the agent has a candidate and wants the detail before committing.

compare

Rank 2–5 packages head-to-head by health.

• Input: a list of 2–5 package names.
• Output: a ranked comparison across the scoring axes.

compare drizzle-orm prisma typeorm

Use this when the choice is already narrowed to a shortlist.

verify

The anti-hallucination guard: is this package real, healthy, and not risky?

• Input: one package name (possibly a misspelling or invention).
• Output: whether it exists, plus a health/risk read.

verify lodahs → not a real package (did you mean lodash?)

Use this before the agent writes an import for a package it "remembers."

diagram

A reference-architecture Mermaid diagram for a chosen stack.

• Input: a description of the stack.
• Output: a Mermaid diagram you can drop into docs or a README.

This one is optional — it's about communicating a stack, not picking packages.

Confidence labels

Every recommendation carries one of four confidence labels, reflecting the strength of the discovered evidence — not anyone's opinion:

See How it works for how these are derived.