pip install doc-distillation-mcp

# With dev tools (MCP Inspector, testing, linting)
pip install 'doc-distillation-mcp[dev]'

# Direct run
doc-distillation-mcp

# Or with uvx (no install needed)
uvx doc-distillation-mcp

# Debug with MCP Inspector
mcp dev doc_distillation_mcp.server:mcp

pip install pdfplumber

{
  "mcpServers": {
    "doc-distillation": {
      "command": "uvx",
      "args": ["doc-distillation-mcp"]
    }
  }
}

{
  "mcpServers": {
    "doc-distillation": {
      "command": "uvx",
      "args": ["doc-distillation-mcp"]
    }
  }
}

{
  "mcpServers": {
    "doc-distillation": {
      "command": "python3",
      "args": ["-m", "doc_distillation_mcp.server"]
    }
  }
}

claude mcp add doc-distillation -- uvx doc-distillation-mcp

# Webpage (sync mode - direct result)
distill_url(url="https://example.com/article")

# With Obsidian subdirectory
distill_url(
    url="https://example.com/deep-dive",
    obsidian_subdir="飞书蒸馏"
)

# Large document (async mode - returns task_id)
distill_url(
    url="https://example.com/long-report.pdf",
    async_mode=True
)
# Then poll:
get_distill_status(task_id="abc12345")

# Text file (sync mode)
distill_file(file_path="/path/to/notes.txt")

# PDF file with Obsidian subdirectory
distill_file(
    file_path="/path/to/report.pdf",
    obsidian_subdir="PDF蒸馏"
)

# Large file (async mode)
distill_file(
    file_path="/path/to/large.pdf",
    async_mode=True
)

get_distill_status(task_id="abc12345")
# Returns: {status: "completed", progress: 1.0, result: {...}}

list_distillations()
# Returns: [{task_id, title, source_type, method, segment_count, ...}]

URL / File Input
    │
    ├─ Stage 1: Source Detection & Content Extraction
    │   ├─ Feishu: Returns guidance (use lark-doc skill)
    │   ├─ Webpage: HTML parsing (text, headings, images)
    │   ├─ PDF: pdfplumber text extraction
    │   ├─ Video/Podcast: Returns guidance (use video-transcript-mcp)
    │   └─ Local file: Type-based extraction
    │
    ├─ Stage 2: Integrity Safeguard
    │   ├─ Structure skeleton (heading hierarchy)
    │   └─ Key element detection (8 categories)
    │
    ├─ Stage 2.5: Image Filtering (three-layer)
    │   ├─ Layer 1: Automatic rules (size, URL keywords, duplicates, alt keywords)
    │   ├─ Layer 2: Context prediction (nearby text indicates value)
    │   └─ Layer 3: Safety net (near key elements)
    │
    ├─ Stage 3: HTML Distillation Article Generation
    │   └─ Styled HTML with header, content sections, footer
    │
    └─ Stage 4: Obsidian Note Generation
        ├─ Frontmatter (title, source, author, date, tags)
        ├─ Summary callout
        ├─ Content sections
        ├─ Image distillation callouts
        └─ Key element statistics table