@dicepdf/mcp

Local-first PDF tools for AI agents — a Model Context Protocol server powered by @dicepdf/core.

One promise: the agent works on files that never leave the machine. Unlike cloud PDF APIs, DicePDF processes everything in-process on the user's own disk — no uploads, no third-party servers, no egress.

Why

PDF wrangling (extract text, merge, split, redact metadata) is one of the most common things an agent needs to do. Every other option uploads the user's private documents to a remote API. This one doesn't.

Install into Claude Desktop / Cursor / Cline

Add to your MCP config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "dicepdf": {
      "command": "npx",
      "args": ["-y", "@dicepdf/mcp"]
    }
  }
}

Then ask your agent things like "extract the text from ~/Downloads/report.pdf" or "merge every PDF in ~/invoices into one file."

Tools

| Tool | Does | |------|------| | pdf_extract_text | Extract text as plain text / Markdown / JSON | | pdf_merge | Merge PDFs in order | | pdf_split | Split by ranges / pages / every page | | pdf_rotate | Rotate pages 90/180/270° | | pdf_delete_pages | Remove pages | | pdf_compress | Structural compression (aggressive rasterizes scans) | | pdf_strip_metadata | Remove Title/Author/Producer/XMP/ID | | pdf_watermark | Stamp a text watermark | | pdf_add_page_numbers | Add page numbers | | pdf_to_image | Render pages to PNG/JPEG files | | pdf_metadata | Page count / form / signature inspection |

Every tool takes local file paths and writes outputs back to disk.

License

MIT