percollate-mcp
# percollate-mcp
An MCP server that wraps [Percollate](https://github.com/danburzo/percollate) to convert web pages into PDF, EPUB, HTML, or Markdown — directly from your AI assistant.
## Tools
### Download tools
Each writes a file to disk:
| Tool | Output |
|------|--------|
| `percollate_download_pdf` | PDF |
| `percollate_download_epub` | EPUB e-book |
| `percollate_download_html` | HTML file |
| `percollate_download_md` | Markdown file |
**Parameters:**
- `urls` (required) — Array of URLs to convert
- `output` — Output file path
- `title` — Document title
- `author` — Document author
- `css` — Additional CSS overrides
- `style` — Custom CSS stylesheet path
- `template` — Custom HTML template path
- `hyphenate` — Enable hyphenation
- `individual` — Export each URL as a separate file
- `wait` — Seconds between processing URLs
- `no_amp` — Don't prefer AMP versions
- `cover` — Generate a cover page
- `toc` — Generate a table of contents
- `toc_level` — Heading depth for ToC (1–6)
- `inline` — Embed images as base64
- `debug` — Verbose output
### percollate_read_md
Reads a web page as Markdown and returns the content directly in conversation. Results are cached locally — repeated calls for the same URL return instantly.
**Parameters:**
- `url` (required) — URL to read
- `title`, `author`, `css`, `style`, `template`, `hyphenate`, `wait`, `no_amp`, `debug` — standard processing options
Cached files are stored in your OS temp directory (`percollate-mcp-cache/`).
## Quick Start
Add to your MCP client config (e.g. Claude Desktop `claude_desktop_config.json`):
```json
{
"mcpServers": {
"percollate": {
"command": "npx",
"args": ["-y", "percollate-mcp"]
}
}
}
```
No install needed — `npx` handles it automatically.
## Requirements
- Node.js 18+
## License
MIT
TDQS
Scored across 5 tools
Each download tool targets a distinct output format, and read_md is separated by its return-content behavior. However, download_md and read_md both produce Markdown, so an agent could momentarily confuse saving a file with reading content, though the descriptions clarify the difference.
All tools use a consistent snake_case naming pattern with the same percollate_ prefix and clear action_format structure. The slight semantic switch from download_* to read_md is still predictable and readable.
Five tools is well-scoped for a web-page conversion server. Each tool maps to a meaningful output or consumption mode without redundant entries.
The set covers the core percollate output formats (PDF, EPUB, HTML, Markdown) plus direct Markdown reading. No obvious lifecycle gaps exist for a stateless conversion/read server.