Document Converter (MarkItDown)
Convert Office and structured documents to Markdown for LLM consumption using Microsoft MarkItDown.
Overview
MarkItDown is a Microsoft open-source tool that converts a wide range of document formats into clean Markdown text. This is essential for academic research workflows where source materials arrive in diverse formats (PDFs from journals, PPTX from conferences, DOCX from collaborators, XLSX data tables).
Supported Formats
| Format |
Extensions |
Notes |
| PDF |
.pdf |
Text extraction; scanned PDFs require OCR |
| Word |
.docx |
Paragraphs, tables, lists, headings preserved |
| PowerPoint |
.pptx |
Slide-by-slide with speaker notes |
| Excel |
.xlsx, .xls |
Sheet-by-sheet, tables as Markdown tables |
| HTML |
.html, .htm |
Cleaned content extraction |
| CSV |
.csv |
Converted to Markdown table |
| JSON |
.json |
Pretty-printed structured output |
| XML |
.xml |
Structured text extraction |
| Images |
.jpg, .png, .gif, .bmp, .tiff |
OCR text extraction (requires optional deps) |
| Audio |
.mp3, .wav |
Transcription (requires optional deps) |
| ZIP |
.zip |
Recursively converts contained files |
MCP Tool Usage
When the markitdown-mcp server is configured, use the convert_to_markdown tool:
Tool: convert_to_markdown
Parameters:
uri: "file:///absolute/path/to/document.pptx"
The uri parameter accepts:
- Local files:
file:///path/to/file.docx
- HTTP URLs:
https://example.com/paper.pdf
Important
- Always use absolute paths with the
file:// URI scheme for local files.
- The tool returns the full Markdown text content of the converted document.
- For large documents, the output may be substantial. Consider summarizing or extracting relevant sections after conversion.
CLI Fallback
If the MCP server is not available, use the CLI directly via shell:
# Convert a single file
markitdown document.pptx
# Save output to file
markitdown document.pptx > output.md
# Convert from URL
markitdown https://example.com/paper.pdf
Academic Research Workflows
1. Conference Presentation Analysis
Convert a PPTX to Markdown, then analyze the content:
- Convert the presentation with
convert_to_markdown
- The output preserves slide structure and speaker notes
- Summarize key findings, methodology, and conclusions
2. Paper Review from PDF
- Convert the PDF to Markdown
- Extract sections: Abstract, Introduction, Methods, Results, Discussion
- Cross-reference with literature library
3. Data Table Extraction
- Convert XLSX to Markdown tables
- Each sheet becomes a separate section
- Use the extracted tables for analysis or comparison
4. Batch Literature Processing
When processing multiple documents:
- Convert each document individually
- Extract metadata (title, authors, date) from the content
- Add to the literature library with
library_add_paper
Output Quality Notes
| Format |
Quality |
Caveats |
| DOCX |
Excellent |
Complex layouts may lose formatting |
| PPTX |
Good |
Diagrams/charts become text descriptions |
| XLSX |
Good |
Merged cells may not render perfectly |
| PDF |
Variable |
Depends on PDF type (text vs scanned) |
| HTML |
Good |
JavaScript-rendered content not captured |
| CSV |
Excellent |
Direct table conversion |
When to Use This Skill
- User uploads or references an Office document (PPTX, DOCX, XLSX)
- User wants to analyze content from a PDF
- User needs to extract text from images (OCR)
- User wants to convert a document for inclusion in notes or workspace
- User asks to read, summarize, or analyze a non-Markdown file
- Batch processing of research materials in mixed formats
When NOT to Use
- File is already Markdown or plain text
- User wants to edit the original document format (use appropriate editor)
- User needs pixel-perfect layout preservation (use the original format)
- File is a scanned PDF without OCR support configured
Troubleshooting
| Issue |
Solution |
markitdown not found |
Run pip install 'markitdown[all]' |
| MCP tool not available |
Ensure markitdown-mcp is installed and configured in openclaw.json |
| Empty output from PDF |
PDF may be scanned/image-only; needs OCR dependencies |
| Encoding errors |
Ensure file is not corrupted; try re-saving from source application |
| Large file timeout |
Convert via CLI instead of MCP for very large files |
Installation
Local (macOS/Linux)
pip install 'markitdown[all]' markitdown-mcp
Verify
markitdown --help
Docker
MarkItDown is pre-installed in the Research-Claw Docker image.
1---2name: document-converter3description: Convert Office documents (PPTX, DOCX, XLSX, PDF, HTML, CSV, JSON, XML, images) to Markdown using Microsoft MarkItDown. Provides the agent with conversion strategies for academic and research workflows.4---5
6# Document Converter (MarkItDown)
7
8Convert Office and structured documents to Markdown for LLM consumption using Microsoft MarkItDown.
9
10---
11
12## Overview
13
14MarkItDown is a Microsoft open-source tool that converts a wide range of document formats into clean Markdown text. This is essential for academic research workflows where source materials arrive in diverse formats (PDFs from journals, PPTX from conferences, DOCX from collaborators, XLSX data tables).
15
16---
17
18## Supported Formats
19
20| Format | Extensions | Notes |
21|--------|-----------|-------|
22| PDF | `.pdf` | Text extraction; scanned PDFs require OCR |
23| Word | `.docx` | Paragraphs, tables, lists, headings preserved |
24| PowerPoint | `.pptx` | Slide-by-slide with speaker notes |
25| Excel | `.xlsx`, `.xls` | Sheet-by-sheet, tables as Markdown tables |
26| HTML | `.html`, `.htm` | Cleaned content extraction |
27| CSV | `.csv` | Converted to Markdown table |
28| JSON | `.json` | Pretty-printed structured output |
29| XML | `.xml` | Structured text extraction |
30| Images | `.jpg`, `.png`, `.gif`, `.bmp`, `.tiff` | OCR text extraction (requires optional deps) |
31| Audio | `.mp3`, `.wav` | Transcription (requires optional deps) |
32| ZIP | `.zip` | Recursively converts contained files |
33
34---
35
36## MCP Tool Usage
37
38When the `markitdown-mcp` server is configured, use the `convert_to_markdown` tool:
39
40```
41Tool: convert_to_markdown
42Parameters:
43 uri: "file:///absolute/path/to/document.pptx"
44```
45
46The `uri` parameter accepts:
47- Local files: `file:///path/to/file.docx`
48- HTTP URLs: `https://example.com/paper.pdf`
49
50### Important
51
52- Always use absolute paths with the `file://` URI scheme for local files.
53- The tool returns the full Markdown text content of the converted document.
54- For large documents, the output may be substantial. Consider summarizing or extracting relevant sections after conversion.
55
56---
57
58## CLI Fallback
59
60If the MCP server is not available, use the CLI directly via shell:
61
62```bash
63# Convert a single file
64markitdown document.pptx
65
66# Save output to file
67markitdown document.pptx > output.md
68
69# Convert from URL
70markitdown https://example.com/paper.pdf
71```
72
73---
74
75## Academic Research Workflows
76
77### 1. Conference Presentation Analysis
78
79Convert a PPTX to Markdown, then analyze the content:
80
811. Convert the presentation with `convert_to_markdown`
822. The output preserves slide structure and speaker notes
833. Summarize key findings, methodology, and conclusions
84
85### 2. Paper Review from PDF
86
871. Convert the PDF to Markdown
882. Extract sections: Abstract, Introduction, Methods, Results, Discussion
893. Cross-reference with literature library
90
91### 3. Data Table Extraction
92
931. Convert XLSX to Markdown tables
942. Each sheet becomes a separate section
953. Use the extracted tables for analysis or comparison
96
97### 4. Batch Literature Processing
98
99When processing multiple documents:
1001. Convert each document individually
1012. Extract metadata (title, authors, date) from the content
1023. Add to the literature library with `library_add_paper`
103
104---
105
106## Output Quality Notes
107
108| Format | Quality | Caveats |
109|--------|---------|---------|
110| DOCX | Excellent | Complex layouts may lose formatting |
111| PPTX | Good | Diagrams/charts become text descriptions |
112| XLSX | Good | Merged cells may not render perfectly |
113| PDF | Variable | Depends on PDF type (text vs scanned) |
114| HTML | Good | JavaScript-rendered content not captured |
115| CSV | Excellent | Direct table conversion |
116
117---
118
119## When to Use This Skill
120
121- User uploads or references an Office document (PPTX, DOCX, XLSX)
122- User wants to analyze content from a PDF
123- User needs to extract text from images (OCR)
124- User wants to convert a document for inclusion in notes or workspace
125- User asks to read, summarize, or analyze a non-Markdown file
126- Batch processing of research materials in mixed formats
127
128---
129
130## When NOT to Use
131
132- File is already Markdown or plain text
133- User wants to edit the original document format (use appropriate editor)
134- User needs pixel-perfect layout preservation (use the original format)
135- File is a scanned PDF without OCR support configured
136
137---
138
139## Troubleshooting
140
141| Issue | Solution |
142|-------|----------|
143| `markitdown` not found | Run `pip install 'markitdown[all]'` |
144| MCP tool not available | Ensure `markitdown-mcp` is installed and configured in `openclaw.json` |
145| Empty output from PDF | PDF may be scanned/image-only; needs OCR dependencies |
146| Encoding errors | Ensure file is not corrupted; try re-saving from source application |
147| Large file timeout | Convert via CLI instead of MCP for very large files |
148
149---
150
151## Installation
152
153### Local (macOS/Linux)
154
155```bash
156pip install 'markitdown[all]' markitdown-mcp
157```
158
159### Verify
160
161```bash
162markitdown --help
163```
164
165### Docker
166
167MarkItDown is pre-installed in the Research-Claw Docker image.
wentorai