Docs Fetch MCP Server

A Bun-based Model Context Protocol (MCP) server for fetching documentation pages and bounded documentation crawls.

The server exposes one MCP tool, fetch_doc_content. It fetches a root URL, extracts readable Markdown, ranks links, and can crawl linked pages within explicit depth, page, timeout, and scope limits. Results are returned as structured JSON with crawl metadata, page content, ranked links, truncation flags, and per-page errors.

Features

• Fast static fetch path with axios
• Optional Puppeteer rendering for client-rendered or thin pages
• Markdown extraction with headings, lists, code blocks, tables, blockquotes, and links
• Real crawl depth semantics:
  • depth: 1 returns only the root page
  • depth: 2 includes direct child links
  • depth: 3 includes grandchildren, up to the maximum of 5
• URL normalization before dedupe: fragments, tracking params, default ports, and trailing slash duplicates are removed
• Crawl scoping by same origin and optional pathPrefix
• Bounded maxPages, maxConcurrency, global timeout, per-page timeout, and per-page content truncation
• Partial results with explicit per-page failures
• Runtime argument validation shared with the advertised tool schema

Requirements

• Bun >=1.3.0
• Puppeteer's browser installation if render is auto or always

This project uses Bun for dependency management, runtime execution, builds, and tests.

Installation

bun install
bun run build

Configure your MCP client:

{
  "mcpServers": {
    "docs-fetch": {
      "command": "bun",
      "args": ["/path/to/docs-fetch-mcp/build/index.js"]
    }
  }
}

For local development without a build:

{
  "mcpServers": {
    "docs-fetch": {
      "command": "bun",
      "args": ["/path/to/docs-fetch-mcp/src/index.ts"]
    }
  }
}

Tool

fetch_doc_content

Fetch one URL and optionally crawl linked pages.

Parameters:

Name	Type	Default	Limit	Description
url	string	required	HTTP/HTTPS only	Root URL to fetch.
depth	number	1	1 to 5	Link distance from the root.
maxPages	number	10	1 to 50	Maximum pages returned across the crawl.
maxConcurrency	number	3	1 to 8	Maximum pages fetched at once.
timeoutMs	number	45000	5000 to 120000	Global crawl timeout.
perPageTimeoutMs	number	10000	1000 to 60000	Per-page fetch timeout.
render	string	auto	auto, always, never	Browser rendering strategy.
sameOrigin	boolean	true	n/a	Restrict crawled links to the same origin.
pathPrefix	string	omitted	n/a	Optional path prefix crawl scope, such as /docs.
contentLimit	number	12000	1000 to 50000	Markdown characters per page before truncation.
includeLinks	boolean	true	n/a	Include ranked links in returned page objects.

Example request:

{
  "url": "https://example.com/docs",
  "depth": 2,
  "maxPages": 8,
  "pathPrefix": "/docs",
  "render": "auto"
}

Response shape:

{
  "rootUrl": "https://example.com/docs",
  "normalizedRootUrl": "https://example.com/docs",
  "explorationDepth": 2,
  "maxPages": 8,
  "pagesExplored": 3,
  "pagesFailed": 1,
  "timedOut": false,
  "durationMs": 1234,
  "crawl": {
    "sameOrigin": true,
    "pathPrefix": "/docs",
    "maxConcurrency": 3,
    "render": "auto",
    "perPageTimeoutMs": 10000
  },
  "content": [
    {
      "url": "https://example.com/docs",
      "finalUrl": "https://example.com/docs",
      "depth": 0,
      "status": 200,
      "title": "Documentation",
      "description": "Example documentation",
      "canonicalUrl": "https://example.com/docs",
      "headings": ["Documentation"],
      "content": "# Documentation\n\n...",
      "contentLength": 2400,
      "truncated": false,
      "fetchedWith": "http",
      "links": [
        {
          "url": "https://example.com/docs/api",
          "text": "API reference",
          "score": 18.5,
          "internal": true
        }
      ]
    }
  ],
  "errors": [
    {
      "url": "https://example.com/docs/missing",
      "depth": 1,
      "error": "Request failed with status code 404",
      "status": 404
    }
  ]
}

Crawl Behavior

• Crawling is breadth-first.
• URLs are normalized before dedupe and enqueue.
• sameOrigin: true rejects links whose origin differs from the normalized root URL.
• pathPrefix further restricts links to a path prefix on the root origin.
• includeLinks: false hides links in returned page objects but does not disable link discovery for crawling.
• render: "never" uses only the HTTP fetch path.
• render: "always" uses Puppeteer for every fetched page.
• render: "auto" tries HTTP first, then uses Puppeteer when HTTP fails or extracted content is very thin.
• Browser fallback currently preserves the rendered response status and extracts the page body even for non-2xx pages.

Architecture

src/index.ts                         CLI entrypoint
src/server.ts                        MCP server wiring and tool registration
src/config/tool-options.ts           Shared tool schema/default/range metadata
src/tool/fetch-doc-content-args.ts   Runtime argument validation
src/crawler/docs-crawler.ts          Crawl orchestration and queue management
src/crawler/page-fetcher.ts          HTTP fetch, browser fallback, extraction coordination
src/browser/browser-manager.ts       Reusable Puppeteer browser/page handling
src/content/content-extractor.ts     Page metadata/content extraction facade
src/content/main-content-selector.ts Main content selection
src/content/link-extractor.ts        Link normalization, dedupe, and ranking
src/content/markdown-renderer.ts     HTML-to-Markdown rendering
src/content/text-cleanup.ts          Shared text normalization helpers
src/utils/url.ts                     URL normalization and scope utilities
src/types/index.ts                   Shared TypeScript types

The MCP schema and runtime option normalization share the same metadata source in src/config/tool-options.ts. Keep new options there first, then wire behavior through validation and crawler options.

Development

bun install
bun run dev
bun run test
bun run typecheck
bun run build

Scripts:

• bun run dev: run the MCP server from TypeScript source.
• bun run test: run Bun tests under src.
• bun run typecheck: run TypeScript with --noEmit.
• bun run build: emit build/index.js with a Bun shebang.
• bun run start: run the built MCP server.

Testing

Tests are colocated with the modules they cover:

• src/crawler/docs-crawler.test.ts: crawl depth, scoping, failures, and link visibility.
• src/crawler/page-fetcher.test.ts: fetch/render fallback characterization.
• src/content/content-extractor.test.ts: Markdown extraction and truncation.
• src/tool/fetch-doc-content-args.test.ts: boundary validation and schema/default sync.
• src/utils/url.test.ts: URL normalization and scope helpers.

When changing behavior, add or update characterization tests first. For refactors, keep bun run test, bun run typecheck, and bun run build green.

Notes

• Use render: "never" for fast static documentation crawls and tests.
• Use pathPrefix for documentation sites that share a domain with marketing pages, blogs, or apps.
• Puppeteer browser installation can be skipped only if callers use render: "never".
• The custom Markdown renderer is intentionally covered by characterization tests; preserve output compatibility unless making an explicit behavior change.

License

MIT