HTML to PDF from Claude and LLM agents (MCP)

pdfrender ships a built-in MCP server: agents call the render_html_to_pdf tool to produce real PDFs. Setup for Claude Code and any MCP client.

MCP endpoint	https://api.pdfrender.dev/mcp/ (HTTP transport)
Tool	render_html_to_pdf — returns { pdf_base64, pages }; input capped at 256 KB
Auth	X-API-Key header optional — anonymous works at the per-IP rate limit; same credits as REST

A PDF is something no LLM can produce locally — pdfrender exposes its renderer as a Model Context Protocol server, so agents generate documents as a tool call. Add it to Claude Code:

claude mcp add --transport http pdfrender https://api.pdfrender.dev/mcp/ \
  --header "X-API-Key: YOUR_KEY"

Or in a generic MCP client configuration:

{
  "mcpServers": {
    "pdfrender": {
      "type": "http",
      "url": "https://api.pdfrender.dev/mcp/",
      "headers": { "X-API-Key": "YOUR_KEY" }
    }
  }
}

The agent then calls render_html_to_pdf(html) and receives {"pdf_base64": "…", "pages": N} — decode the base64 and write the file.

Tool vs REST

The tool mirrors the REST policy — anonymous calls work at the per-IP limit, a key raises it, one call costs one credit — but caps input at 256 KB (base64 megabytes inside tool results are hostile to agent context windows). For bigger documents have the agent call POST /v1/render instead. The same rendering rules apply: HTML + CSS, data: URIs, no JavaScript.

Try it free

Try it free — 100 renders a month, no card — the free tier covers an agent's occasional document easily.