pick_library
Given a coding task, recommend 1-3 libraries from the should-i-use index with one-line reasoning and an honest warning about the top pick. Use before installing anything.
MCP serveropen source992 libraries live
should-i-use - the MCP that tells your agent what NOT to install
Honest MCP library recommendations before your coding agent adds another dependency.
Claude Codeone-line install
claude mcp add should-i-use -- npx -y should-i-use-mcpMeasured against Context7
A repeatable benchmark compared task-focused answers from should-i-use with the calls an agent makes to retrieve Context7 documentation.
Test10 rounds x 20 queries
should-i-use median238 tokens
Context7 median1,154 tokens
Measured result4.8x vs Context7
Randomly sampled libraries, measured across two days. Read the method and every raw row in the GitHub bench/ folder.
Six tools, each built for a decision
should_i_use
Honest verdict on a specific library: 4-axis scores with reasons, when to skip it, and maintenance signals (last push, weekly downloads).
alternatives
Curated alternatives to a library, each with a one-line "prefer it when" and a verdict if the alternative is also indexed.
how_do_i
Return the 1-2 best code snippets for a task with a library, plus the gotcha for each. Snippets are correct for the indexed version. Far cheaper in tokens than a docs dump.
audit_dependencies
Audit a project's dependencies for unmaintained packages, unstable APIs, and documented reasons to switch. Pass the contents of package.json, requirements.txt, or pyproject.toml. Returns only the dependencies worth a decision, not a report on every line. Run this before working in an unfamiliar codebase.
docs_link
Official docs URL and GitHub repo for a library, one line each. Use when the clamped answers are not enough.
Honest limitations
Worth knowing before you rely on it:
Coverage is curated, not complete. An audit stays quiet on unindexed packages, so a clean result means only that nothing indexed was flagged.
The tools are advisory. An agent decides when to call MCP tools unless you add the optional project policy with npx should-i-use-mcp install --with-policy.
Guides are research, not test drives. They use documentation, release notes, changelogs, issue history, and a fixed rubric. Popular entries also receive clean-container install checks, but every release is not installed.
Guides are point-in-time. Signals refresh daily and guides move through a review pipeline, so a new release or maintainer change can arrive before the written verdict catches up.
pick_library ranks on keywords, not embeddings. Plain descriptions work best; unusual phrasing may need a rewrite.
The benchmark is indicative, not a general law. It sampled libraries at random across two days, and its method and raw rows are public.
Verdicts are informed opinions. They are sourced and argued, but disagreeing with one is a valid bug report.
Run it locally or use the hosted endpoint
The npm package runs locally with no API key, database, or network access at runtime. The index ships with the package.
A hosted streamable HTTP endpoint is rolling out at mrkeyoor.com/mcp. It uses the live index; the npx install above is the stable path today.
FAQ
What is should-i-use?
should-i-use is an MCP server that gives coding agents concise library recommendations, skip warnings, alternatives, maintenance signals, and task-focused code snippets.
How is should-i-use different from Context7?
Context7 retrieves library documentation. should-i-use is built for dependency decisions: what to pick, what to skip, why, and which alternative fits better, with compact responses that protect the context budget.
Is should-i-use free?
Yes. The MCP server is free to install from npm, its code is MIT licensed, and the published library data is available under CC BY 4.0.
How current is the data?
Registry and repository signals refresh daily. Guides are reviewed on a pipeline and remain point-in-time research, so check the official docs link when a recent release may have changed the answer.
What if a verdict is wrong?
Open a GitHub issue with the library, the verdict you dispute, and current evidence. Challenges and corrections are part of how the index improves.
Open an issue on GitHub