v0.10.0

v0.10.0 Highlights

• Bounded multi-shelf discovery for open-source projects and competing commercial products.
• Dual-box evidence separation (Projects you could reuse vs Products you would compete with).
• Lean, keyless default operation with optional GitHub token and Tavily web discovery.

v0.2.3

Now listed in the official MCP registry as io.github.aradar46/reuse-before-generate.

claude mcp add reuse-before-generate -- npx -y reuse-before-generate

Changes since v0.2.2

• Added server.json and the mcpName field so the package can be published to the official MCP registry.
• No functional changes. Search, filtering and output are identical to v0.2.2.

See v0.2.2 for what the server actually does.

v0.2.2

First public release.

An MCP server that checks whether your idea already exists before your
agent builds it. Searches GitHub, npm, and Python repos, drops anything
abandoned, and hands the surviving candidates to your agent to judge.

claude mcp add reuse-before-generate -- npx -y reuse-before-generate

What's in it

• One tool, check_before_building, that returns maintained alternatives
  plus scoring instructions — the calling agent does the semantic judgment
  using the session already running, so no separate API key is needed.
• Search across three lanes: GitHub relevance, a stars:0..3 lane for
  small or brand-new repos that GitHub's ranking buries, and a
  language:python lane.
• A "maintained" filter based on recent activity, deliberately not on
  star count — a 0-star repo doing exactly your job is the duplicate most
  worth catching.
• Partial failures are named in the output, so fewer results are never
  silently mistaken for a complete search.

Measured, not asserted

71 offline tests, plus a scored recall eval over 12 known cases:

Target found in the top 10	11 of 11
Found in the top 5	10 of 11
False positives on the true-negative case	0

Methodology and known gaps: docs/findings.md.

Requires Node 18+. MIT.