docling.rs — Claude Code MCP server: what it does, how to install
docling.rs MCP server converts PDF, Office, HTML and scans to Markdown with structure intact — headings, tables and OCR for agents.
claude mcp add --env DOCLING_RS_CACHE_DIR=<DOCLING_RS_CACHE_DIR> --env DOCLING_RS_OCR_LANG=<DOCLING_RS_OCR_LANG> --transport stdio doclingrs -- uvx docling-rs-mcpWhat docling.rs does
docling.rs parses documents and returns content agents can read: Markdown with headings in order, tables cell by cell, scanned pages through OCR. It runs layout analysis models, TableFormer for table recognition and Whisper for audio, all from a single Rust binary without Python or PyTorch.
Handles PDF, Office (Word, Excel, PowerPoint), HTML, email, images and audio. PDF converts 4.3× faster than Python docling, Office and HTML 20–60× faster. Table cell recall 0.94, MarkItDown gets 0.31. Office, HTML and Markdown need no models; for PDF and images the server downloads ONNX models on first run (under 1 GB).
Outputs Markdown, JSON, HTML, LaTeX. Reads URLs, data: URIs and YouTube (captions). Works over stdio, connects to any MCP-capable platform.
What it helps with
- Extract text from PDF preserving heading structure and tables
- Run OCR on scans and images to pull out text
- Convert Office documents to Markdown for agent processing
- Transcribe audio and video to text with Whisper
- Parse HTML and email into structured format
Example prompts for the agent
- Convert
report.pdfto Markdown and show tables - Extract text from scan
scan.pngand save tooutput.md - Parse all PDFs in
./docsfolder and output heading structure
How to install docling.rs
Platform CLI
claude mcp add --env DOCLING_RS_CACHE_DIR=<DOCLING_RS_CACHE_DIR> --env DOCLING_RS_OCR_LANG=<DOCLING_RS_OCR_LANG> --transport stdio doclingrs -- uvx docling-rs-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from docling-rs-mcp <command>
Platform CLI
codex mcp add doclingrs --env DOCLING_RS_CACHE_DIR=<DOCLING_RS_CACHE_DIR> --env DOCLING_RS_OCR_LANG=<DOCLING_RS_OCR_LANG> -- uvx docling-rs-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from docling-rs-mcp <command>
Config file
{
"mcpServers": {
"doclingrs": {
"command": "uvx",
"args": [
"docling-rs-mcp"
],
"env": {
"DOCLING_RS_CACHE_DIR": "<DOCLING_RS_CACHE_DIR>",
"DOCLING_RS_OCR_LANG": "<DOCLING_RS_OCR_LANG>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from docling-rs-mcp <command>
Config file
{
"servers": {
"doclingrs": {
"type": "stdio",
"command": "uvx",
"args": [
"docling-rs-mcp"
],
"env": {
"DOCLING_RS_CACHE_DIR": "<DOCLING_RS_CACHE_DIR>",
"DOCLING_RS_OCR_LANG": "<DOCLING_RS_OCR_LANG>"
}
}
}
}Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from docling-rs-mcp <command>
Platform CLI
gemini mcp add -e DOCLING_RS_CACHE_DIR=<DOCLING_RS_CACHE_DIR> -e DOCLING_RS_OCR_LANG=<DOCLING_RS_OCR_LANG> doclingrs uvx docling-rs-mcpReplace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from docling-rs-mcp <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about docling.rs
How to install docling.rs in Claude Code?
Run `claude mcp add --env DOCLING_RS_CACHE_DIR=<DOCLING_RS_CACHE_DIR> --env DOCLING_RS_OCR_LANG=<DOCLING_RS_OCR_LANG> --transport stdio doclingrs -- uvx docling-rs-mcp`. The server downloads models for PDF and OCR on first use.
What does docling.rs do?
Converts documents — PDF, Office, HTML, scans, audio — to Markdown and JSON with structure preserved: headings, tables, OCR. Works as an MCP server, uses layout analysis models and TableFormer.
Which formats does docling.rs support?
PDF, Word, Excel, PowerPoint, HTML, Markdown, email, images (PNG, JPEG) and audio. Office and HTML need no models; for PDF and images the server downloads ONNX models on first use.
