Claude Tooling

A public notebook of one repeated question:

Does a Claude Code plugin, MCP server or skill already exist for this — or do I have to build it?

Every answer gets written down, dated, and left here. Not because the answers are interesting on their own, but because the search is the expensive part and it is otherwise thrown away the moment the decision is made.

Why this exists

I build a fair number of plugins and MCP servers. The default failure mode when you can build things quickly is building something that already exists, badly. The opposite failure mode — spending an afternoon proving a negative and then repeating that afternoon four months later because nobody wrote it down — is just as expensive and much less obvious.

So the rule here is:

• Don't reinvent the wheel. If someone has built it and it fits, use theirs.
• Don't build what someone else has built better. A partial fit from a maintained project usually beats a perfect fit I have to maintain alone.
• But do build when the requirement is genuinely unmet. Specific requirements that nothing serves are the legitimate case, and they're more common than they should be.

A scan closes with one of those three, in writing, against criteria written before the searching started.

Why it's public

Two reasons, and the second is the real one.

1. Anyone wondering the same thing can read the answer instead of redoing the search.
2. Agents can find it. A negative result — "I looked in these six places on this date and there is nothing" — is real information that no search engine produces on its own, because nobody publishes the search that came up empty. Leaving it here turns a dead end into a breadcrumb.

Nothing in this repo is private by nature. Scans describe requirements and public software; where a requirement comes from something of mine that isn't public, it gets stated as a capability, not as a description of the thing.

How to read a scan

Scans live in scans/, one file per question, indexed in scans/README.md.

Each has a verdict — one of a fixed set, defined in docs/verdicts.md:

Verdict	Meaning
already-have-it	I'd already built it and forgotten
native	Claude Code or the API does this now; no extension needed
adopt	Something exists, it fits, use it
adopt-with-gaps	Something exists and mostly fits; the gaps are listed
contribute	Closest match is alive and short of the spec — send a PR
fork	Closest match is abandoned — fork it rather than start over
wrong-shape	Something serves the need, but not as a plugin — an MCP already on my gateway, a CLI, ten lines of shell
build	Nothing meets the spec. Blue sky

Every scan is a snapshot with a date on it. This ecosystem moves in weeks. A build verdict from six months ago is a hypothesis, not a fact — see the Revisions section at the bottom of each scan, and rescan.

Running one

From inside this repo:

/scan a plugin that scrapes Amazon listings and checks geo-specific delivery options

Or just describe the need — the prior-art-scan skill triggers on it. The procedure is in .claude/skills/prior-art-scan/SKILL.md; the search surfaces, with working queries and their known blind spots, are in docs/method.md.

The output is a report in scans/, an updated index, and a one-line recommendation.

Ship public, use private

When a scan ends in build, the thing that gets built is public by default. That is possible because a plugin doesn't have to choose between being shareable and being useful to me specifically — it can declare where its user data lives and let each user's copy point somewhere different:

${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/<plugin-name>/

The plugin is public and identical for everyone; my instance reads my directory. No parallel private fork of the same idea, no divergence between the copy that works and the copy that's published.

That convention, and the three-lifecycle architecture it belongs to, are documented properly in Plugin–Workspace–Vault rather than restated here.

What this is not

Three related things of mine answer neighbouring questions. Sending a request to the wrong one wastes a scan:

Question	Goes to
I have a requirement — does anything already serve it?	here
I have a domain (diagramming, OCR) — what software in it is worth wrapping?	create-claude-plugins:plugin-target-research
I have a specific tool — is it worth wrapping?	create-claude-plugins:assess-plugin-candidate
Nothing exists — should I build a skill, a plugin, an MCP or a hook?	claude-rudder:what-thing-pick

This repo hands off to the last one whenever a scan returns build.

Licence

MIT — see LICENSE.