wellworn
MCP server · works without a key

Know what breaks before you build.

Wellworn is the judgment layer for coding agents. Before your agent adds a dependency or lays out a screen, it asks which one, what breaks at that version, and who verified it.

claude mcp add --transport http wellworn https://mcp.wellworn.dev/mcp
Works in
Browse verdicts →
One call from the agent, one answer back: the pick and why, the alternative and when it wins, what to avoid, the traps at that version. The last line carries the reviewer, the date, and the day it has to be checked again.
01

The agent asks before it chooses

One tool call, recommend, compare, alternatives, traps, design, or docs, at the moment a dependency or a screen is about to be added. The Claude Code plugin runs it from a hook on package install commands and on edits to a dependency manifest.

02

It gets a verdict, not a page

A pick, one alternative and the condition under which it wins, what to avoid, the traps at that version, and the skill to do it right. One block, capped at 3,200 characters, with the traps trimmed first when it does not fit.

03

Every line carries its date

Who verified it, against which version, and when it expires. A nightly job re-reads the registries; when a major version moves, the verdict is served with a RECHECK flag instead of going quietly stale.

04

Your traps feed the next agent

Hit a breakage, fix it, and submit_trap sends the symptom and the fix to a reviewer. Approved traps are served with your name on the provenance line.

What it is wired to

Four corpora behind the same tool call.

Design systems

Which system to build a screen in, the tokens it ships, and the components that are not there yet.

22 systems indexed
Skills

The instruction file that goes with the pick, so the agent writes the library the way its maintainers do.

none published yet
Open-source catalog

Self-hosted tools with stars, archive state, licence, and platform, so the alternatives tool answers from data.

catalog: 28,242 tools
Docs pointers

The official docs URL and llms.txt for a package, so the agent reads the source instead of guessing.

7 llms.txt on file
Browse the catalog

The categories people ask about most.

All categories

Pricing

Free to use every month. Paid when it is doing real work.

Caps are per organization, not per key. Answers are small, so the free tier costs little to run and stays free.

No key
free

60 calls a day per IP. Every read tool.

No account needed
Free key
free

2,000 calls a month per organization. Submit traps.

Create a key →
Pro
$12 / month

20,000 calls a month, credit packs past the cap, priority verdict requests.

See plans →
Team
$49 / month

100,000 calls pooled, 5 seats, private notes on verdicts.

See plans →
Questions people ask first

A README tells you what the maintainer wants to be true today. A verdict tells you what a named reviewer observed at a specific version on a specific date, with the sources, and it expires. When the two disagree, the verdict page shows the dispute button.

Add it to your agent in one line.

No account, no key. Sixty calls a day to see whether it changes what your agent picks.

claude mcp add --transport http wellworn https://mcp.wellworn.dev/mcp