PolyDoc for AI agents (MCP)

Model Context Protocol (MCP) is an open standard that lets AI assistants and agents call external tools directly. polydoc-mcp is an MCP server that exposes PolyDoc's PDF, screenshot, and e-invoice generation as tools an agent can call, so it can produce a real, downloadable file instead of describing one.

Installation

Requires Node.js 20.15 or later. The server is published to npm and runs with npx, no install step required:

npx -y polydoc-mcp

Configuration

The server is configured entirely through environment variables.

Variable	Required	Default	Description
POLYDOC_API_KEY	yes	-	Your API key from dashboard.polydoc.tech.
POLYDOC_SANDBOX	no	false	Default sandbox mode (watermarked output, does not use production quota). A tool call can override this per request.
POLYDOC_BASE_URL	no	https://api.polydoc.tech	Override for self-hosted or staging use.
POLYDOC_OUTPUT_DIR	no	OS temp dir + /polydoc	Where downloaded files are written. Output never escapes this folder.

Claude Desktop

Add to claude_desktop_config.json (Settings > Developer > Edit Config):

{
  "mcpServers": {
    "polydoc": {
      "command": "npx",
      "args": ["-y", "polydoc-mcp"],
      "env": {
        "POLYDOC_API_KEY": "your-key"
      }
    }
  }
}

Claude Code

claude mcp add polydoc -e POLYDOC_API_KEY=your-key -- npx -y polydoc-mcp

Cursor

Add the same mcpServers block shown above to ~/.cursor/mcp.json, or to a project-level .cursor/mcp.json.

Available tools

Tool	What it does
polydoc_html_to_pdf	Renders a URL, raw HTML, or a saved template to PDF. Supports layout, margins, page format, page ranges, bookmarks, and accessible/tagged PDF output.
polydoc_screenshot	Renders a URL, raw HTML, or a saved template to PNG, JPEG, or WebP, with viewport and device-pixel-ratio control. Returns an inline image preview when small enough for the client to display.
polydoc_generate_einvoice	Generates a Factur-X or ZUGFeRD hybrid PDF/A-3 from structured invoice data, with profiles from minimum to extended.
polydoc_test_credentials	Verifies the configured API key with a minimal sandbox render. Never uses production quota.

Content for the PDF and screenshot tools can come from a URL, an inline HTML string, or a saved template (with Liquid templateData), exactly as in the REST API. Every tool also accepts an advanced object that is merged into the request body, so any API capability not exposed as a typed field (for example pdf.watermark, pdf.encryption, or render options) is still reachable. See the full field reference for everything the underlying API accepts.

Output and delivery

• Download (default): the file is written under POLYDOC_OUTPUT_DIR and the tool returns the absolute path plus metadata (size, conversion id, credits used). Screenshots also return an inline image block. Pass returnBase64: true to receive the raw bytes as base64 instead, for clients without filesystem access.
• Cloud storage: set delivery: "cloudStorage" with a presignedUrl to have PolyDoc upload directly to your own S3, GCS, Azure Blob, Backblaze B2, or Wasabi bucket. See the cloud storage guide for the presigned-URL setup on each provider.
• Webhook: set delivery: "webhook" with a webhook object ({ url, async?, method?, headers? }) for HMAC-signed asynchronous delivery.

What an agent can do with this

Once configured, you can ask an agent things like:

• "Render this HTML as a PDF and save it to my Downloads folder."
• "Take a full-page screenshot of https://example.com at 2x device pixel ratio."
• "Generate a PDF from my invoice template with this data, tagged as accessible PDF/UA."
• "Check that my PolyDoc API key works before I run the rest of this workflow."

Compliance output (PDF/A archival, PDF/UA-1 accessible/tagged PDF) works the same way through MCP as through the REST API: it is best-effort, and the optional veraPDF verification step still applies. See the PDF/A and PDF/UA guides for exact parameters. A request with pdf.pdfua.verify: true (or the PDF/A equivalent) that fails conformance returns an error instead of a passing file, whether the call came from an agent or a direct API call.

Source

polydoc-mcp is MIT-licensed and open source, with worked examples for each tool in the repository's examples/ folder.