Rossum MCP & Agent
rossumaiLABSCO SUMMARY
The format list is the product.
Almost every extraction tool covers PDF and Word. The reason to reach for this one is the long tail — Hangul documents, signed email, AppleWorks and Lotus files, images that need OCR — where the alternative is a manual conversion step. The trade is that documents go to the Unstructured API, so it is the wrong choice for anything you cannot send off the machine.
What it is
An MCP server that hands a model the contents of a document it otherwise cannot open. It calls the Unstructured partitioning API, so the format coverage is unusually wide: alongside the obvious PDF and Office files it handles email, ebooks, images, Hangul and Korean office formats, spreadsheets, and older formats like .wk2 and .dbf. The author labels the repository work in progress.
What you get
- Text and structure extracted from a document so the model can reason over its content rather than its filename
- Documents:
.pdf,.doc,.docx,.odt,.rtf,.pages,.epub,.txt,.md,.rst,.org,.html,.htm,.xml - Spreadsheets and data files:
.xls,.xlsx,.xlsm,.xlsb,.csv,.tsv,.dif,.dbf,.fods - Presentations:
.ppt,.pptx,.pptm,.pot,.potm - Email and signed mail:
.eml,.p7s - Images that need reading rather than viewing:
.png,.jpg,.jpeg,.tiff,.bmp,.heic,.gif,.webp,.svg - Legacy and regional formats most extractors drop:
.hwp,.uof,.uos1,.uos2,.abw,.cwk,.mcw,.sdp,.sgl,.wk2
Requirements
An Unstructured API key in UNSTRUCTURED_API_KEY, placed in a .env in the project root — extraction happens on Unstructured's side, so document content leaves your machine and usage bills to that account. Clone the repository, set up the uv environment, and run uv run doc_processor.py; the client config points at the project directory with absolute paths. The README is explicit that this is work in progress.
Setup effort
Build from source — git clone https://github.com/MKhalusova/unstructured-mcp.git
USE CASES
01
Asking questions of a PDF the model cannot open
Extract the contents first so the answer comes from the document's text rather than from its title.
02
Reading a format nothing else handles
Point it at the Hangul document, the signed email or the old spreadsheet instead of hunting for a converter.
03
Getting text out of a scanned page
Hand it the image file directly — TIFF, HEIC, PNG — and work with what comes back.
BEFORE YOU INSTALL
What to know first
Requires
UNSTRUCTURED_API_KEY in your environment
Where it runs
The documented client runs the server on your machine.
INSTALLATION
Connect your client
Server setup
TERMINAL
git clone https://github.com/MKhalusova/unstructured-mcp.gitCreate a .env file in the root directory and add the following env variable: UNSTRUCTURED_API_KEY.
MCP Clients
Claude Desktop
CLAUDE DESKTOP · JSON
{
"mcpServers": {
"unstructured_doc_processor": {
"command": "<YOUR_PATH_TO_UV>",
"args": [
"--directory",
"<YOUR_PATH_TO_UNSTRUCTURED_MCP>",
"run",
"doc_processor.py"
],
"disabled": false
}
}
}The configuration on this page does not set UNSTRUCTURED_API_KEY; the project's instructions put it in a .env file. Check which folder the server reads that file from before you rely on it.
