> [DomPruner] stripe.com
> | Raw HTML  | 305,778 tokens |
> | Refined   |     988 tokens |
> | Reduction |        99.7%   |
> Fetch: 1,546ms · Parse: 8.2ms

npx -y dompruner-mcp

{
  "mcpServers": {
    "dompruner": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "dompruner-mcp"]
    }
  }
}

{
  "mcpServers": {
    "dompruner": {
      "command": "npx",
      "args": ["-y", "dompruner-mcp"]
    }
  }
}

{
  "mcpServers": {
    "dompruner": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "dompruner-mcp"]
    }
  }
}

dompruner_fetch(url?, query?)

> [DomPruner] fastapi.tiangolo.com
> |           | Tokens |
> |-----------|--------|
> | Raw HTML  | 35,905 |
> | Refined   |  2,011 |
> | Reduction |  94.4% |
> Fetch: 83ms · Parse: 20.7ms

# First Steps

The simplest FastAPI file could look like this:
...

dompruner_analyze(url)

URL
 └─▶ fetchPage()       — tiered fetch: direct → UA rotation → Playwright fallback
      │
      ├─▶ [SSG]  extractSsgMarkdown()
      │           walks __NEXT_DATA__ RSC tree → clean Markdown  (≥ 90% reduction)
      │           skips DOM parse entirely
      │
      └─▶ [SSR/CSR]  pre-strip <script>/<style>/<svg>  (80–92% size ↓)
                └─▶ parse5 DOM Tree
                     └─▶ FQN Router (L1)        keeps p / h1–h5 / li / pre / code
                          │                      prunes nav / footer / aside / form
                          └─▶ Heading Cluster (L2)  dev-doc structure detection
                               └─▶ CETD Engine (L3)  text-density scoring fallback
                                    └─▶ BM25+ Section Filter  query-aware ranking
                                         └─▶ Semantic Anchor  heading hierarchy + meta
                                              └─▶ Compact Markdown  ──▶  LLM context

npm install playwright-core
npx playwright install chromium

src/
  mcp-server.ts          — MCP stdio transport + tool/prompt handlers
  pipeline.ts            — Orchestrator: fetch → parse → extract → rules → serialize
                           Includes 5-min TTL URL cache
  ast/
    fetcher.ts           — Tiered HTTP fetch (native → UA rotation → Playwright)
    parser.ts            — Pre-strip (<script>/<style>/<svg>) + parse5 DOM builder
    core-extractor.ts    — L1→L2→L3 extraction cascade
    fqn-router.ts        — L1: FQN semantic selector matching + noise pruning
    heading-cluster.ts   — L2: heading-block clustering for developer docs
    cetd.ts              — L3: Content/Tag-Density scoring fallback
    ssg-extractor.ts     — Next.js / Nuxt / Gatsby __NEXT_DATA__ RSC tree walk
    anchor.ts            — Semantic anchor extraction (title, meta description, h1–h3)
  middleware/
    serializer.ts        — FQNNode[] → Compact Markdown + token estimator
  rule-engine/
    registry.ts          — URL pattern → Rule set resolution and chaining
    types.ts             — Rule interface definitions
    builtin/
      section-bm25.ts    — BM25+ with heading boost + ancestor preservation
      http-endpoint.ts   — HTTP method/path/response block reformatter
      code-signature.ts  — Code block function/class signature extractor

git clone https://github.com/dong7812/AST-RAG-MCP.git
cd AST-RAG-MCP
npm install
npm run dev    # tsx watch — no build step needed
npm run build  # tsc → dist/

echo '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"dompruner_fetch","arguments":{"url":"https://fastapi.tiangolo.com","query":"routing"}}}' \
  | npm run dev 2>/dev/null