file extraction
Extract Office and spreadsheet files to Markdown.
Give your agent Word docs, Excel workbooks, CSV files, TSV exports, or PowerPoint decks. Frenchie turns them into Markdown your agent can read, cite, and work with.
$Extract ./q3-forecast.xlsx to Markdown with Frenchie
# Extracted Markdown ## Sheet: Summary | Field | Value | | ----- | ----- | | ARR | $4.9M | | Risk | Procurement delay | ## Notes - Source file preserved as agent-readable Markdown. - Your agent does the reasoning after extraction.
why it exists
Files first. Reasoning after.
Agents are good at reasoning over text. They are less good at opening a workbook, preserving sheet boundaries, keeping CSV tables sane, or pulling slide content into a shape that fits a prompt. That gap usually turns into brittle scripts and one-off converters.
Frenchie returns clean Markdown, not a schema, not a summary, not a document-intelligence platform. Your agent still does the reasoning. Frenchie handles the file-reading layer.
formats
One extraction tool. Five file shapes.
Word to Markdown for AI Agents
Contracts, briefs, policies, and reports land as DOCX all the time. Frenchie gives your agent Markdown instead of asking it to guess through a binary file.
Read the format guide →Excel to Markdown for AI Agents
Frenchie turns XLSX workbooks into sheet-by-sheet Markdown so your agent can inspect assumptions, compare tabs, and write the follow-up without a custom parser.
Read the format guide →CSV to Markdown for AI Agents
CSV is simple until your agent has to reason over it. Frenchie handles quoting, headers, truncation, and output shape so the agent gets a clean table.
Read the format guide →TSV to Markdown for AI Agents
TSV files show up in data exports, logs, and research workflows. Frenchie turns them into Markdown tables without asking your agent to parse tabs by hand.
Read the format guide →PowerPoint to Markdown for AI Agents
Frenchie extracts PowerPoint decks slide by slide, so your agent can write a summary, briefing note, or follow-up without opening presentation software.
Read the format guide →quick start
Install skills and MCP tools together.
The installer configures the MCP server and installs the Frenchie skills alongside the MCP tools, so your agent has the callable tools and the instructions for when to use them.
Install Frenchie
$ npx @lab94/frenchie install --api-key fr_...
This writes the local stdio MCP config and copies the Frenchie skills into your project.
Ask your agent to extract
> Extract ./q3-forecast.xlsx to Markdown with Frenchie
Your agent calls extract_to_markdown. Stdio results auto-save to .frenchie/<name>/result.md.
Reason over Markdown
DOCX costs 0.5 credit per page. XLSX costs 0.5 credit per sheet. CSV and TSV cost 0.5 credit per file. PPTX costs 1 credit per slide. DOCX, XLSX, and PPTX files can be up to 50 MB. CSV and TSV files can be up to 20 MB. Results expire 30 minutes after first delivery.
workflow
How it runs.
- 1
Install Frenchie in your agent. The command writes the MCP config and copies the skills together.
- 2
Drop in a DOCX, XLSX, CSV, TSV, or PPTX file and ask your agent to extract it.
- 3
Your agent calls `extract_to_markdown`. Small jobs usually return inside the smart-wait window; larger jobs return a `jobId` and finish async.
- 4
The Markdown lands at `.frenchie/<name>/result.md` in stdio mode. HTTP agents upload first, then persist the final Markdown locally.
questions
The practical bits.
What file types does extract_to_markdown support?
DOCX, XLSX, CSV, TSV, and PPTX. PDFs and images still go through `ocr_to_markdown`; audio and video still go through `transcribe_to_markdown`.
Is this structured data extraction?
No. Frenchie converts files into clean Markdown. If you need fields, JSON, classifications, or summaries, your agent does that after reading the Markdown.
Where does the result go?
In stdio mode, Frenchie auto-saves `.frenchie/<name>/result.md` next to your work. In HTTP mode, the agent uploads first and must save the final Markdown locally before the task is complete.
What happens to my source file?
Frenchie processes the file and deletes it. Result payloads expire 30 minutes after first delivery.
Give your agent a real file.
Install Frenchie, drop in a DOCX, XLSX, CSV, TSV, or PPTX file, and see the Markdown your agent gets back.