Convert raw HTML strings to clean markdown via html-to-markdown-mcp (npx)

Recipe: Raw HTML → clean markdown via html-to-markdown-mcp

Launch

npx -y html-to-markdown-mcp

Transport: stdio (NDJSON). Server: html-to-markdown v1.0.0. Powered by Turndown.js.

Tool inventory

Verified trace

1. Convert complex HTML with tables, lists, code blocks (id:3)

→ {"method":"tools/call","params":{"name":"html_to_markdown","arguments":{"html":"<h1>Agent Registry</h1><p>The <strong>tani</strong> registry indexes tools, APIs and MCP servers.</p><ul><li>Ranked by <em>invocation trust</em></li><li>Verified by execution</li><li>Content-addressed responses</li></ul><table><tr><th>Surface</th><th>Kind</th><th>Trust</th></tr><tr><td>mcp-server-fetch</td><td>MCP</td><td>100</td></tr><tr><td>mcp-server-git</td><td>MCP</td><td>100</td></tr></table><blockquote>One real verified recipe beats ten plausible fakes.</blockquote><pre><code class=\"language-json\">{\"protocol\":\"tani/1.0\",\"op\":\"resolve\"}</code></pre>"}}}

← Output:

# Agent Registry

**Source:** Unknown
**Saved:** 2026-06-13T04:13:32.743Z

---

# Agent Registry

The **tani** registry indexes tools, APIs and MCP servers.

-   Ranked by _invocation trust_
-   Verified by execution
-   Content-addressed responses

Surface Kind Trust
mcp-server-fetch MCP 100
mcp-server-git MCP 100

> One real verified recipe beats ten plausible fakes.

```json
{"protocol":"tani/1.0","op":"resolve"}

Key differences from mcp-server-fetch

• Accepts raw HTML strings directly (html param) — no URL needed
• Use when you already have HTML from another source (API response, DOM extraction, email body)
• Also supports URL fetching as alternative input (url param)
• maxLength param truncates output for large pages
• saveToFile saves full content to disk and returns a summary
• Adds metadata header (source, timestamp) by default

Quirks

• Table rendering loses column alignment — outputs as flat text, not markdown tables. Tables with complex formatting may need post-processing.
• Metadata header is included by default; set includeMetadata: false to skip.
• Heading from the HTML is duplicated (once in metadata, once in body).