dsh-feed — Cross-ecosystem aggregation base ("aggregation of aggregations")

It does not directly build UI; it provides an open data layer: normalizes the GitHub dsh-plugin topic and npm registry into an open JSON index for any Agent / CLI / MCP / future UI to consume. Your plugin only does "sync + query" — others build UI on top of this data.

Why it's needed

• Existing marketplaces each scrape data and maintain it themselves, closed-source and duplicative. dsh-feed becomes the data source: sync once, queryable across the entire ecosystem.

• One plugin, three consumption surfaces:

  • Model tools: feed_sync / feed_search / feed_stats

  • CLI: dsh-feed sync|search|stats|mcp

  • MCP server (experimental): any MCP client (Cursor, Claude Desktop…) can query it

• The index is an open JSON file (default .dsh/dsh-feed.json), plain text, diffable, and readable by any tool.

Related MCP server: npm-search-mcp-server

Installation

dsh plugin --profile <profile> add dsh-feed
# 或只用 CLI（npm 全局）：npm i -g dsh-feed

Usage

Model side

CLI

dsh-feed sync                      # 同步（可选 GITHUB_TOKEN 提升限额）
dsh-feed search "移动端 远程"      # 查询
dsh-feed stats
dsh-feed mcp                       # 极简 stdio MCP server（实验性）

MCP configuration example (any MCP client)

{ "mcpServers": { "dsh-feed": { "command": "dsh-feed", "args": ["mcp"] } } }

Index format (open data layer)

{
  "updated": "2026-08-16T...",
  "sources": { "github": 200, "npm": 180 },
  "count": 380,
  "plugins": [
    { "name": "owner/repo", "url": "...", "description": "...", "stars": 42, "topics": ["dsh-plugin", "notify"], "npmName": "dsh-xxx" }
  ]
}

Design

• lib/sync.js — scraping + normalization (GitHub search API + npm search API, token injectable).

• lib/query.js — pure query logic (tokenization/scoring/statistics), zero-dependency and unit-testable.

• lib/index-store.js — index file read/write (relative path + anti-escape validation).

• bin/dsh-feed.js — CLI + minimal stdio MCP server (JSON-RPC 2.0).

• Security: The index is written only to .dsh/ inside the session workspace; network requests happen only in feed_sync (explicitly triggered).

Roadmap

• Hourly scheduled sync (dsh-schedule integration).

• Web API read-only endpoints + more complete MCP tools (pagination/filtering/details).

• Share the same data layer with dsh-need-finder / dsh-recipe (deduplicate scraping).

Testing

node test/feed.test.mjs

License

MIT