OpenDocKit
Read local PDF, Office, HWP, OpenDocument, RTF, EPUB, HTML and email files as Markdown. Offline.
At a glance
Not tested yet- GitHub stars
- 41
- Weekly downloads
- 1,570
- Last commit
- 2026-11-21
- Licence
- MIT
- Transport
- stdio
- Auth
- none
uvx opendockit-mcp
What it does
OpenDocKit is a document converter that turns files into clean Markdown for language models. It reads PDF, DOCX, PPTX, XLSX, HWP/HWPX, ODF, RTF, EPUB, HTML, email and images. It is built for the files people actually upload, including Korean HWP documents and messy tables. It is for developers and agent users who need documents in a form an LLM can read.

Key features
- Keeps structure, not just text: merged table cells, multi-column PDFs, borderless tables, nested lists, footnotes, speaker notes, charts as data tables and equations as LaTeX.
- Reads Korean documents, including a from-scratch HWP 5.x parser, HWPX, CP949 detection and NFC normalization.
- Detects file type by content rather than extension, and warns when it does.
- Bounds risky uploads such as zip bombs and recursive PDF objects, and can run each file in a disposable worker process with memory and time limits.
- Extracts images in place and can describe them with a vision model if you plug one in.
- Stops work at a character budget instead of truncating afterwards.
- MIT licensed, with no ML models and no GPU for plain conversion.
Quick start
- Install LibreOffice if you need to read legacy
.docand.pptfiles. - Start the MCP server with a root folder:
opendockit-mcp --root ~/Documents
- Add the server to your client's MCP settings so the client can start it.
Add this to .mcp.json in your project root (Claude Code) or .cursor/mcp.json (Cursor):
{
"mcpServers": {
"opendockit": {
"command": "uvx",
"args": [
"opendockit-mcp"
],
"env": {
"OPENDOCKIT_MCP_ROOTS": "YOUR_OPENDOCKIT_MCP_ROOTS"
}
}
}
}Replace each YOUR_ value with your own, and delete the variables you do not use.
Environment variables
OPENDOCKIT_MCP_ROOTS: the sources do not explain this variable.
Example prompts
- Read the quarterly report PDF in my Documents folder and summarize it.
- Convert this Word file to Markdown and keep the tables.
- Open the HWP file and tell me what page 3 says.
- List the slides in this PowerPoint and their speaker notes.
What to watch out for
- Requires Python 3.12 or newer on Linux, macOS or Windows.
- On Windows, a malformed HTML page or mail body over 256 K characters, or one nested very deeply, is read by an older parser and may lose content, with a warning.
- Legacy
.docand.pptfiles need LibreOffice installed. - The
heicextra pulls in GPL/LGPL-licensed codec libraries. - Image description needs a vision model and reads
OPENAI_API_KEY.
Not tested by tools4agents yet. Read the source repository before you install it.
Similar tools
Atomic Mail
Programmable email inbox for AI agents, JMAP, PoW auth, stdio MCP server.
Smitline: AI Phone Calls & Meetings
Let AI agents make phone calls and join Zoom, Teams and Google Meet meetings. Hosted or self-hosted.
Slack
Read channels, search message history, and post updates, so an agent can pull context from where decisions actually happened.
IMAP/SMTP email for AI agents, read, send, organize, and manage attachments across accounts.