
Distill Mcp
io.github.lakshgkPublished by GitHub account lakshgkv0.1.1Updated Oct 24, 2026
Convert PDF, DOCX, XLSX, PPTX, HTML, audio to token-efficient Markdown for any MCP client.
Overview
Converts local documents such as PDF, DOCX, XLSX, PPTX, HTML and audio into token-efficient Markdown for an MCP client.
- What you can do
- Distill Mcp exposes a convert_and_save tool that turns local files into clean Markdown before the model reads them, with typical token reductions of 40 to 80 percent versus raw document text. It handles Word, Excel, PowerPoint, native-text PDF, HTML, EPUB, JSON, SQL and WSDL files, and in full mode also scanned PDFs via OCR and audio formats. Converted output is cached, and the tool reports whether a previous cached version was overwritten and any warnings.
- When to use it
- Useful when an assistant needs to read or reason over local documents and context budget matters, for example summarizing reports, spreadsheets or slide decks. It fits workflows where the client is instructed to convert files first instead of reading originals through filesystem tools.
- Requirements
- Runs locally over stdio, launched with npx -y distill-mcp-server, so Node.js is needed. Lightweight mode additionally needs Python and pip, configured through the DISTILL_MCP_CONFIG environment variable with a python_path. Full mode needs Docker with the Distill service running locally, configured through DISTILL_MCP_CONFIG with a distill_url. No accounts or API keys are declared.
Installation
In SourceWeft
- Open Distill Mcp in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
distill-mcp
[npm version] [license] [MCP Registry]
An MCP server that connects any MCP-compatible client to Distill, converting local documents to clean, token-efficient Markdown before the LLM reads them. Typical token reduction is 40–80% compared to raw document text, letting the model fit more content into its context window and reason over it faster.
Works with Claude Desktop, Claude Code, Cursor, Windsurf, and any other MCP client that supports tool calling.
Two Modes
Lightweight needs only Python and pip. Full needs Docker with the Distill service running. See the setup guides for details.
Quick install
No global install required:
Then configure Claude Desktop — see Lightweight setup or Full setup for step-by-step instructions.
Configuration
Add the server to your claude_desktop_config.json.
Lightweight mode — macOS / Linux:
Lightweight mode — Windows:
Full mode — all platforms:
Full config key reference is in the setup guides: Lightweight | Full
CLAUDE.md snippet
Paste this into your project's CLAUDE.md so Claude knows how to use the
tool automatically. Copy the block below as-is:
The snippet is also available in docs/CLAUDE.md-snippet.md.
Supported formats
Privacy
All processing happens locally. In lightweight mode, documents are converted by the distill-core Python library on your machine — no data leaves your computer. In full mode, documents are sent to the Distill Docker service running locally on your machine — no data is sent to external services.
Built on
Distill — document-to-Markdown conversion engine.
License
MIT
Source: README.md at commit da01804
Tools
0Version history
1- v0.1.1LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts