Why? AI agents generate data, but users need real documents — spreadsheets with charts, formatted Word reports, branded slide decks. Doc Hive bridges the gap.

Contents

• Features
• Companion: LINE Hive
• Quick Start
• How It Works
• Prerequisites
• Installation
• MCP Tools — XLSX · DOCX · PPTX · Templates · Inspect
• PPTX Theme System — Presets · Org Templates · Per-slide Directives
• Environment Variables
• Development
• Architecture
• Contributing
• License

Features

• 📊 Excel (XLSX) — multi-sheet workbooks from CSV with column styles, formulas, merges, frozen headers
• 📝 Word (DOCX) — documents from Markdown with headings, lists, tables, images, page setup
• 📽️ PowerPoint (PPTX) — presentations from Marp-style Markdown with auto-detected layouts and 4 built-in presets
• 🎨 Theme system — gradient backgrounds, title bars, logos, footers, slide numbers
• 🏢 Org templates — branded JSON configs for company-wide consistency
• 🐍 Python bridge — fill existing .pptx templates via python-pptx
• 📎 File output — all documents written to a configurable output directory with absolute paths

Companion: LINE Hive

Doc Hive generates documents — LINE Hive delivers them. Together, your agents can create polished spreadsheets and slide decks, then send them straight to your phone via LINE.

%%{init: {"theme": "default"}}%%
flowchart TB
    subgraph Generate["📄 Doc Hive"]
        direction LR
        XLSX["create_xlsx"] ~~~ PPTX["create_pptx"]
    end
    subgraph Deliver["🐝 LINE Hive"]
        direction LR
        Ask["line_ask"] ~~~ Send["line_send_message"]
    end
    Agent["🤖 Agent"] --> Generate
    Generate -.-> File["💾 Files"]
    File -.-> Deliver
    Agent --> Deliver
    Deliver --> Phone["📱 LINE"]

    style Generate fill:#e3f2fd,stroke:#1565c0
    style Deliver fill:#f3e5f5,stroke:#7b1fa2
    classDef output fill:#e8f5e9,stroke:#2e7d32
    class Phone output

Doc Hive works standalone — no LINE Hive required. But pairing them lets agents generate and deliver in one workflow. See the LINE Hive README for full documentation.

Quick Start

npm install @zoulabo/doc-hive
npm install exceljs docx pptxgenjs   # document libraries you need
npx doc-hive init                    # interactive setup
# Restart your editor — MCP server starts automatically

Or set up for a specific editor directly:

npx doc-hive init --editor cursor    # or: vscode, claude-code, claude-desktop

Tell your agent: "Create a spreadsheet with Q1 revenue data" — it handles the rest.

How It Works

Agents call MCP tools with structured data. Doc Hive generates polished files and returns the path.

sequenceDiagram
    participant Agent as 🤖 Agent
    participant Hive as 📄 Doc Hive
    participant FS as 💾 File System

    Agent->>Hive: create_xlsx({ title, csv, columns })
    Hive->>FS: Write .xlsx
    Hive-->>Agent: { filePath: "/out/Report.xlsx" }

    Agent->>Hive: create_pptx({ markdown, theme: "dark" })
    Hive->>FS: Write .pptx
    Hive-->>Agent: { filePath: "/out/Deck.pptx" }

    Agent->>Hive: inspect_template({ template: "corp.pptx" })
    Hive-->>Agent: { slides, placeholders, layouts }
    Agent->>Hive: fill_template({ template, data })
    Hive->>FS: Write filled .pptx
    Hive-->>Agent: { filePath: "/out/Filled.pptx" }

Key concepts:

• Structured input — CSV for spreadsheets, Markdown for docs and slides, JSON for templates
• Zero config — sensible defaults; customize with themes, presets, or org templates
• Optional deps — only install the document libraries you actually use

Prerequisites

• Node.js ≥ 18
• At least one document library:

| Feature | Package | Install | |---------|---------|---------| | Excel (XLSX) | exceljs | npm install exceljs | | Word (DOCX) | docx | npm install docx | | PowerPoint (PPTX) | pptxgenjs | npm install pptxgenjs | | PPTX templates | python-pptx | pip install python-pptx |

Python is only needed for fill_template and inspect_template. All other tools use pure Node.js.

Installation

npm install @zoulabo/doc-hive
npx doc-hive init                    # interactive setup

The wizard detects your editor and writes the right config. You can also specify the editor directly:

npx doc-hive init --editor cursor    # or: vscode, claude-code, claude-desktop

Claude Code as VS Code/Cursor extension? Use --editor claude-code (not --editor vscode).

Manual Configuration

If you prefer manual setup:

Add to .vscode/mcp.json:

{
  "servers": {
    "doc-hive": {
      "command": "node",
      "args": ["node_modules/@zoulabo/doc-hive/bin/doc-hive.js"],
      "env": {
        "OUTPUT_DIR": ".doc-output"
      }
    }
  }
}

Add to .mcp.json:

{
  "mcpServers": {
    "doc-hive": {
      "command": "node",
      "args": ["node_modules/@zoulabo/doc-hive/bin/doc-hive.js"],
      "env": {
        "OUTPUT_DIR": ".doc-output",
        "TEMPLATE_DIR": "templates"
      }
    }
  }
}

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "doc-hive": {
      "command": "node",
      "args": ["/path/to/your/project/node_modules/@zoulabo/doc-hive/bin/doc-hive.js"],
      "env": {
        "OUTPUT_DIR": ".doc-output"
      }
    }
  }
}

The server uses stdio transport (stdin/stdout for MCP JSON-RPC). Configure your client to run:

node node_modules/@zoulabo/doc-hive/bin/doc-hive.js

With these environment variables:

• OUTPUT_DIR — output directory (default .doc-hive-out)
• TEMPLATE_DIR — template directory (default templates)
• PYTHON_PATH — Python 3 executable (default python3)

MCP Tools

| Tool | Purpose | Requires | |------|---------|----------| | create_xlsx | Generate Excel workbook from CSV data | exceljs | | create_docx | Generate Word document from Markdown | docx | | create_pptx | Generate PowerPoint from Marp-style Markdown | pptxgenjs | | fill_template | Fill .pptx template placeholders with data | python-pptx | | inspect_template | List slide layouts, placeholders, and shapes | python-pptx | | list_templates | List available .pptx templates | — | | check_python | Verify Python 3 + python-pptx availability | — |

create_xlsx

Generate Excel workbooks from CSV data.

title: "Q1 Revenue Report"
csv: "Product,Revenue,Growth\nWidget A,$1200,12%\nWidget B,$3400,28%"
columns: [{ width: 20 }, { width: 15, style: "currency" }, { width: 12, style: "percent" }]

Features: Multi-sheet, multi-table per sheet, column styles (currency/percent/date/number/text), formulas, cell merges, header styles (dark/light/accent), frozen headers, auto-filters, striped rows.

create_docx

Generate Word documents from Markdown.

title: "Status Report"
markdown: "# Project Status\n\n## Summary\n\nThe project is **on track**.\n\n- Task 1: Complete\n- Task 2: In progress"

Features: H1–H6 headings, bold/italic/code, bullet & numbered lists, pipe-delimited tables, images with dimensions, page breaks (---), page setup (orientation, size, margins), headers & footers with page numbers.

create_pptx

Generate PowerPoint presentations from Marp-style Markdown.

title: "Q1 Review"
markdown: "# Q1 Review\nQuarterly Business Review\n\n---\n\n## Revenue\n\n- Total: $4.6M\n- Growth: 18% YoY"
theme: "dark"

Features: Auto-detected layouts (# Title, ## Content, blank, two-column), 4 built-in presets, gradient backgrounds, gradient title bars, slide numbers, logos, footers, per-slide overrides, background images, tables, blockquotes.

See PPTX Theme System for full details.

fill_template

Fill an existing .pptx template with data using python-pptx. Use inspect_template first to discover placeholder indices.

template: "corporate.pptx"
title: "Q1 Report"
data: {
  slides: [
    { slideIndex: 0, placeholders: { 0: "Q1 Report", 1: "January – March 2024" } },
    { slideIndex: 1, placeholders: { 0: "Revenue Summary" }, images: { 13: "chart.png" } }
  ],
  globalReplacements: { "{{company}}": "ACME Corp", "{{year}}": "2024" }
}

inspect_template

List slide layouts, placeholders, and shapes in a .pptx template. Run before fill_template to discover the template structure.

list_templates / check_python

Utility tools — list available .pptx files in the template directory, or verify that Python 3 and python-pptx are installed.

PPTX Theme System

Theme Presets

Pass theme: "dark" (string) or theme: { preset: "dark", ... } (object with overrides).

| Preset | Background | Text | Accent | Title Bar | |--------|-----------|------|--------|-----------| | dark | Gradient navy | White | Purple #6C63FF | Gradient | | corporate | White | Dark | Blue #1a73e8 | Gradient | | vibrant | Gradient warm | White | Orange #ff6b35 | Gradient | | minimal | Light gray | Dark | Gray #666666 | Flat |

Theme Resolution Chain

Per-slide directives (<!-- bg: #hex -->)
  → Per-call overrides ({ theme: { primaryColor: "#ff0000" } })
    → Named org template (template: "corporate" → templates/corporate.json)
      → Workspace default (templates/default.json)
        → Built-in preset (theme: "dark")
          → Hardcoded defaults (purple accent, dark text, Calibri)

Organization Templates

Create a JSON file in your template directory to enforce brand consistency:

// templates/corporate.json
{
  "primaryColor": "#003366",
  "fontColor": "#FFFFFF",
  "fontFamily": "Noto Sans JP",
  "background": { "gradient": ["#003366", "#001a33"] },
  "titleBar": "gradient",
  "slideNumbers": { "format": "{n} / {total}" },
  "logo": {
    "path": "assets/logo.png",
    "position": "top-right",
    "width": 1.2,
    "height": 0.5
  },
  "footer": {
    "text": "© 2024 ACME Corp",
    "position": "bottom-left"
  }
}

Use it: create_pptx({ ..., template: "corporate" }) — Doc Hive reads templates/corporate.json and applies all settings.

Per-slide Directives

Override theme settings on individual slides using HTML comments in Markdown:

| Directive | Effect | |-----------|--------| | <!-- bg: #hex --> | Override slide background color | | <!-- fontColor: #hex --> | Override text color | | <!-- titleBar: none --> | Hide title bar on this slide | | <!-- no-slide-number --> | Hide slide number on this slide | | <!-- logo: none --> | Hide logo on this slide | | <!-- layout: two-column --> | Force two-column layout |

Environment Variables

| Variable | Default | Description | |----------|---------|-------------| | OUTPUT_DIR | .doc-hive-out | Directory for generated files | | TEMPLATE_DIR | templates | Directory for .pptx templates and org template JSON configs | | PYTHON_PATH | python3 | Path to Python 3 executable |

Output

Generated files are written to the output directory (default .doc-hive-out/). Each tool returns:

{
  "success": true,
  "filePath": "/absolute/path/to/.doc-hive-out/Q1-Revenue-Report.xlsx",
  "fileName": "Q1-Revenue-Report.xlsx"
}

Development

git clone https://github.com/zoulabo/doc-hive.git
cd doc-hive
npm install
npm install exceljs docx pptxgenjs   # optional deps for testing
npm run build
npm test                              # 163 tests
cp .vscode/mcp.json.example .vscode/mcp.json   # MCP dev config

| Command | Purpose | |---------|---------| | npm run build | Compile TypeScript | | npm run watch | Watch mode | | npm test | Run tests (vitest) | | npm start | Start standalone server |

Architecture

• Standalone MCP server — stdio transport, no webhook or HTTP server needed
• Optional dependencies — only loaded when tools are called; graceful error if missing
• Python bridge — spawn with stdin JSON pipe for template operations (no temp files)
• Theme resolution — layered cascade: directives → call overrides → org template → preset → defaults
• File output — all documents written to a single configurable directory with absolute path responses

Contributing

Designed for AI-agent-driven development. The repo includes skill files (.github/skills/) that teach agents the codebase. Point your agent at the repo and let it work.

• npm test must pass (163 tests, <3s)
• npx tsc --noEmit must be clean
• Let the agents read skills first, then write the code

License

MIT