PowerPoint Translator using Amazon Bedrock

April 25, 2026 · View on GitHub

A powerful PowerPoint translation tool that leverages Amazon Bedrock models for high-quality translation. This service can be used both as a standalone command-line tool and as a FastMCP (Fast Model Context Protocol) service for integration with AI assistants like Kiro. It translates PowerPoint presentations while preserving formatting and structure.

Features

New features

Source Language Auto-Detect: One-shot LLM detection from the document text; prompt, cache key, and cost estimate all use the detected language. Skips the API entirely if source == target.
Chart Translation: Translates chart titles, axis labels, category names, and series names — without touching numeric data
Translation Cache: Pluggable SQLite / in-memory cache that reuses prior translations across runs; cache key includes model, language, polishing, and glossary
Custom Glossary: External YAML glossary for consistent terminology per language; auto-detected from ./glossary.yaml
Dry-Run / Cost Preview: --dry-run estimates tokens and cost before actually calling Bedrock
Automatic Retry: tenacity-backed exponential backoff on throttling / transient Bedrock errors
Rich Progress: Real-time progress bar with ETA, cache hit rate, and running token/cost totals — per-file slide progress rows in batch mode

Existing features

PowerPoint Translation: Translate text content in PowerPoint presentations
Amazon Bedrock Integration: Uses Amazon Bedrock models for high-quality translation (Claude Opus 4.7, Sonnet 4.6, Nova, Llama 4, and more)
Format Preservation: Maintains original formatting, layouts, and styles
Language-Specific Fonts: Automatically applies appropriate fonts for target languages
Color & Style Preservation: Preserves original text colors and formatting even for untranslated content
Standalone & MCP Support: Use as a command-line tool or integrate with AI assistants via FastMCP
Multiple Languages: Supports translation between various languages (90+ supported)
Batch Processing: Can handle multiple slides, text elements, and entire folders efficiently (parallel workers)
Selective Translation: Translate entire presentations, specific slides, or all files in a folder

Examples

Translation

The PowerPoint Translator maintains the original formatting while accurately translating content:


Original presentation slide in English with complex layout	Same presentation translated to Korean with preserved formatting and layout

Claude Code MCP Examples


Check MCP	MCP example

Usage Examples

🧾 Quick reference: see docs/cheatsheet.md for a compact list of everyday flags.

Translate entire presentation:

uv run ppt-translate translate samples/en.pptx --target-language ko

standalone

Translate specific slides:

uv run ppt-translate translate-slides samples/en.pptx --slides "1,3" --target-language ko

Batch translate all PPT files in a folder:

# Recursive by default — translates all .pptx in samples/ and its subfolders
uv run ppt-translate batch-translate samples/ --target-language ko

# Top level only (skip subfolders)
uv run ppt-translate batch-translate samples/ --target-language ko --no-recursive

# Specify output folder
uv run ppt-translate batch-translate samples/ -t ja -o output/

# Specify number of parallel workers (default: 4)
uv run ppt-translate batch-translate samples/ -t ko -w 4

# Custom output and workers
uv run ppt-translate batch-translate reInvent-2025/ -t ko -o translated/ -w 8

batch-translate

Get slide information:

uv run ppt-translate info samples/en.pptx

Preview cost before translating (dry-run):

# Estimate tokens / cost without calling Bedrock — no file is written
uv run ppt-translate translate samples/en.pptx -t ko --dry-run

dry-run

Translation cache (enabled by default):

# SQLite cache at ~/.ppt-translator/cache.db — translations are reused across runs
uv run ppt-translate translate samples/en.pptx -t ko

# In-memory cache (per-process, no disk writes)
uv run ppt-translate translate samples/en.pptx -t ko --cache-backend memory

# Disable cache entirely
uv run ppt-translate translate samples/en.pptx -t ko --no-cache

# Custom cache path
uv run ppt-translate translate samples/en.pptx -t ko --cache-path /tmp/my-cache.db

The cache key is sha256(source_text + target_language + model_id + polishing + glossary_hash), so translations are automatically reused only when every relevant input matches. Changing the glossary, language, or model invalidates entries as expected.

Custom glossary (YAML):

# Auto-detects ./glossary.yaml in the current directory
uv run ppt-translate translate samples/en.pptx -t ko

# Or point to a specific glossary file
uv run ppt-translate translate samples/en.pptx -t ko -g my-glossary.yaml

Example glossary.yaml:

ko:
  "API": "API"            # src == tgt → keep as-is (do not translate)
  "Foundation Model": "파운데이션 모델"
  "Observability": "Observability"
ja:
  "Cloud": "クラウド"

Skip chart translation (if you want charts untouched):

uv run ppt-translate translate samples/en.pptx -t ko --no-charts

Source language (auto-detected by default):

# Source language is auto-detected on first run (1 extra Bedrock call per PPT)
uv run ppt-translate translate samples/en.pptx -t ko

# Or specify it explicitly to skip detection
uv run ppt-translate translate samples/en.pptx --source-language en -t ko

# Disable auto-detection entirely (let the model infer from context, like before)
uv run ppt-translate translate samples/en.pptx -t ko --no-detect-source

The detected language is folded into the cache key, so the same text under different source languages won't cross-contaminate. If source == target (e.g., translating a Korean deck to Korean) the tool skips Bedrock entirely.

Prerequisites

Python 3.11 or higher
AWS Account with Bedrock access
AWS CLI configured with appropriate credentials
Access to Amazon Bedrock models (e.g., Claude, Nova, etc.)

AWS Credentials Setup

Before using this service, ensure your AWS credentials are properly configured. You have several options:

AWS CLI Configuration (Recommended):
```
aws configure
```
This will prompt you for:
- AWS Access Key ID
- AWS Secret Access Key
- Default region name
- Default output format

AWS Profile Configuration:

aws configure --profile your-profile-name

Environment Variables (if needed):

export AWS_ACCESS_KEY_ID=your_access_key
export AWS_SECRET_ACCESS_KEY=your_secret_key
export AWS_DEFAULT_REGION=us-east-1

IAM Roles (when running on EC2 instances)

The service will automatically use your configured AWS credentials. You can specify which profile to use in the .env file.

Installation

Clone the repository:

git clone https://github.com/daekeun-ml/ppt-translator
cd ppt-translator

Install dependencies using uv (recommended):
```
uv sync
```
Or using pip:
```
pip install -r requirements.txt
```

Set up environment variables: Edit .env file with your configuration:

# AWS Configuration
AWS_REGION=us-east-1
AWS_PROFILE=default

# Translation Configuration
DEFAULT_TARGET_LANGUAGE=ko
BEDROCK_MODEL_ID=global.anthropic.claude-sonnet-4-6

# Translation Settings
MAX_TOKENS=4000
TEMPERATURE=0.1
ENABLE_POLISHING=true
BATCH_SIZE=20
CONTEXT_THRESHOLD=5

# Font Settings by Language
FONT_KOREAN=맑은 고딕
FONT_JAPANESE=Yu Gothic UI
FONT_ENGLISH=Amazon Ember
FONT_CHINESE=Microsoft YaHei
FONT_DEFAULT=Arial

# Debug Settings
DEBUG=false

# Post-processing Settings
ENABLE_TEXT_AUTOFIT=true
TEXT_LENGTH_THRESHOLD=10

Note: AWS credentials (Access Key ID and Secret Access Key) are not needed in the .env file if you have already configured them using aws configure. The service will automatically use your AWS CLI credentials.

Usage

Standalone Command-Line Usage

The PowerPoint Translator can be used directly from the command line using the ppt-translate command:

# Translate entire presentation to Korean
uv run ppt-translate translate samples/en.pptx --target-language ko

# Translate specific slides (individual slides)
uv run ppt-translate translate-slides samples/en.pptx --slides "1,3" --target-language ko

# Translate slide range
uv run ppt-translate translate-slides samples/en.pptx --slides "2-4" --target-language ko

# Translate mixed (individual + range)
uv run ppt-translate translate-slides samples/en.pptx --slides "1,2-4" --target-language ko

# Get slide information and previews
uv run ppt-translate info samples/en.pptx

# Show help
uv run ppt-translate --help
uv run ppt-translate translate --help
uv run ppt-translate translate-slides --help

FastMCP Server Mode (for AI Assistant Integration)

Start the FastMCP server for integration with AI assistants like Kiro:

# Using uv (recommended)
uv run mcp_server.py

# Using python directly
python mcp_server.py

FastMCP Setup

The same server works with any MCP host (Claude Code, Kiro, Kiro CLI, ...). Pick your host below — the JSON schema is identical; only the config file path differs.

Shared server config

Replace /path/to/ppt-translator/ with the actual path to your clone. AWS_* env vars are optional when aws configure is already set up.

{
  "mcpServers": {
    "ppt-translator": {
      "command": "uv",
      "args": [
        "--project", "/path/to/ppt-translator",
        "run", "/path/to/ppt-translator/mcp_server.py"
      ],
      "env": {
        "AWS_REGION": "us-east-1",
        "AWS_PROFILE": "default",
        "BEDROCK_MODEL_ID": "global.anthropic.claude-sonnet-4-6"
      },
      "disabled": false,
      "autoApprove": [
        "translate_powerpoint",
        "get_slide_info",
        "get_slide_preview",
        "translate_specific_slides"
      ]
    }
  }
}

Prefer running with plain python instead of uv? Swap "command": "uv" and remove "--project", ... from args, leaving "args": ["/path/to/ppt-translator/mcp_server.py"].

Claude Code

Claude Code is Anthropic's official CLI. Two ways to register:

Option 1 — claude mcp add (fastest)

# Project-scoped (only when running Claude Code inside this repo)
claude mcp add ppt-translator \
  --scope project \
  -- uv --project /path/to/ppt-translator run /path/to/ppt-translator/mcp_server.py


claude mcp add ppt-translator \
  --scope project \
  -- uv --project /Users/daekeun/Github/ppt-translator run /Users/daekeun/Github/ppt-translator/mcp_server.py

# User-scoped (available in every project)
claude mcp add ppt-translator \
  --scope user \
  -e AWS_REGION=us-east-1 \
  -e AWS_PROFILE=default \
  -e BEDROCK_MODEL_ID=global.anthropic.claude-sonnet-4-6 \
  -- uv --project /path/to/ppt-translator run /path/to/ppt-translator/mcp_server.py

Restart Claude Code, then run /mcp to confirm the server is connected.

Option 2 — config file

Paste the shared JSON into one of:

Project-scoped: .mcp.json at the repo root (commit to share with the team)
User-scoped: ~/.claude.json under the top-level mcpServers key

Troubleshooting

claude --debug prints MCP connection logs at startup.
Test the server standalone: uv run mcp_server.py. If that errors, fix it before retrying the integration.

Kiro

Install Kiro: https://kiro.dev (CLI: https://kiro.dev/cli)
Paste the shared JSON into the appropriate file:
- Kiro (desktop): ~/.kiro/settings/mcp.json
- Kiro CLI (macOS/Linux): ~/.aws/amazonq/mcp.json
- Kiro CLI (Windows): %APPDATA%\amazonq\mcp.json

Using it

Just ask in natural language — the host picks the right tool automatically:

Translate samples/en.pptx to Korean
Batch-translate samples/ into Japanese, dry-run first
Show me what's in slide 3 of en.pptx

Available MCP Tools

The MCP server provides the following tools:

translate_powerpoint: Translate an entire PowerPoint presentation
- Parameters:
  - input_file: Path to the input PowerPoint file (.pptx)
  - target_language: Target language code (default: 'ko')
  - output_file: Path for the translated output file (optional, auto-generated)
  - model_id: Amazon Bedrock model ID (default from BEDROCK_MODEL_ID env)
  - enable_polishing: Enable natural language polishing (default: true)
  - glossary_file: Path to a glossary YAML file (defaults to ./glossary.yaml if present)
  - cache_backend: "sqlite" (default), "memory", or "none"
  - dry_run: If true, estimate cost without translating (default: false)
  - translate_charts: Translate chart titles/axes/categories/series (default: true)
  - source_language: ISO 639-1 source code. Auto-detected if omitted.
  - auto_detect_source: Run 1-shot LLM language detection when source_language is omitted (default: true)
translate_specific_slides: Translate only specific slides in a PowerPoint presentation
- Parameters:
  - input_file: Path to the input PowerPoint file (.pptx)
  - slide_numbers: Comma-separated slide numbers to translate (e.g., "1,3,5" or "2-4,7")
  - target_language: Target language code (default: 'ko')
  - output_file: Path for the translated output file (optional, auto-generated)
  - model_id: Amazon Bedrock model ID (default from BEDROCK_MODEL_ID env)
  - enable_polishing: Enable natural language polishing (default: true)
  - glossary_file, cache_backend, dry_run, translate_charts, source_language, auto_detect_source: same as above
batch_translate_powerpoint: Translate all PowerPoint files in a folder in parallel
- Parameters:
  - input_folder, target_language, output_folder, model_id, enable_polishing
  - recursive, workers (default: 4)
  - glossary_file, cache_backend, cache_path, dry_run, translate_charts, source_language, auto_detect_source
get_slide_info: Get information about slides in a PowerPoint presentation
- Parameters:
  - input_file: Path to the PowerPoint file (.pptx)
- Returns: Overview with slide count and preview of each slide's content
get_slide_preview: Get detailed preview of a specific slide's content
- Parameters:
  - input_file: Path to the PowerPoint file (.pptx)
  - slide_number: Slide number to preview (1-based indexing)
list_supported_languages: List all supported target languages for translation
list_supported_models: List all supported Amazon Bedrock models
get_translation_help: Get help information about using the translator

Configuration

Environment Variables

AWS_REGION: AWS region for Bedrock service (default: us-east-1)
AWS_PROFILE: AWS profile to use (default: default)
DEFAULT_TARGET_LANGUAGE: Default target language for translation (default: ko)
BEDROCK_MODEL_ID: Bedrock model ID for translation (default: global.anthropic.claude-sonnet-4-6)
BEDROCK_MAX_RETRIES: Max retry attempts for transient Bedrock errors (default: 5)
MAX_TOKENS: Maximum tokens for translation requests (default: 4000)
TEMPERATURE: Temperature setting for AI model (default: 0.1)
ENABLE_POLISHING: Enable translation polishing (default: true)
BATCH_SIZE: Number of texts to process in a batch (default: 20)
CONTEXT_THRESHOLD: Number of texts to trigger context-aware translation (default: 5)
CACHE_BACKEND: Translation cache backend — sqlite / memory / none (default: sqlite)
CACHE_PATH: SQLite cache file location (default: ~/.ppt-translator/cache.db)
DEBUG: Enable debug logging (default: false)

Supported Claude Models (Bedrock)

The latest Anthropic models on Amazon Bedrock are registered in Config.SUPPORTED_MODELS and pricing.py. Common IDs you can pass via --model-id or BEDROCK_MODEL_ID:

Model	Global profile	US profile
Claude Opus 4.7	`global.anthropic.claude-opus-4-7`	`us.anthropic.claude-opus-4-7`
Claude Opus 4.6	`global.anthropic.claude-opus-4-6-v1`	`us.anthropic.claude-opus-4-6-v1`
Claude Sonnet 4.6	`global.anthropic.claude-sonnet-4-6`	`us.anthropic.claude-sonnet-4-6`
Claude Opus 4.5	`global.anthropic.claude-opus-4-5-20251101-v1:0`	`us.anthropic.claude-opus-4-5-20251101-v1:0`
Claude Sonnet 4.5	`global.anthropic.claude-sonnet-4-5-20250929-v1:0`	`us.anthropic.claude-sonnet-4-5-20250929-v1:0`
Claude Haiku 4.5	`global.anthropic.claude-haiku-4-5-20251001-v1:0`	`us.anthropic.claude-haiku-4-5-20251001-v1:0`
Claude 3.7 Sonnet	—	`us.anthropic.claude-3-7-sonnet-20250219-v1:0`

Claude Opus 4.6 / 4.7 also support eu., au. (Opus 4.6), and jp. (Opus 4.7) geo-specific inference profiles. Availability depends on your AWS account / region — check the AWS Bedrock console to confirm.

Supported Languages

The service supports translation between major languages including:

English (en)
Korean (ko)
Japanese (ja)
Chinese Simplified (zh)
Chinese Traditional (zh-tw)
Spanish (es)
French (fr)
German (de)
Italian (it)
Portuguese (pt)
Russian (ru)
Arabic (ar)
Hindi (hi)
And many more...

Troubleshooting

Common Issues

AWS Credentials Not Found:
- Ensure AWS credentials are properly configured
- Check AWS CLI configuration: aws configure list
Bedrock Access Denied:
- Verify your AWS account has access to Bedrock
- Check if the specified model is available in your region
FastMCP Connection Issues:
- Verify the path in mcp.json is correct
- Check that Python and dependencies are properly installed
- Review logs in Q Developer for error messages
- Test the server: uv run python mcp_server.py
PowerPoint File Issues:
- Ensure the input file is a valid PowerPoint (.pptx) file
- Check file permissions for both input and output paths
Module Import Errors:
- Use uv run to ensure proper virtual environment activation
- Install dependencies: uv sync

Development

Project Structure

ppt-translator/
├── mcp_server.py                    # FastMCP server implementation
├── main.py                          # Main entry point
├── ppt_translator/                  # Core package
│   ├── __init__.py                  # Package initialization
│   ├── cli.py                       # Command-line interface
│   ├── ppt_handler.py               # PowerPoint processing logic
│   ├── translation_engine.py        # Translation service (cache/glossary/metrics)
│   ├── bedrock_client.py            # Amazon Bedrock client (with retry)
│   ├── retry.py                     # tenacity-based retry policy for Bedrock
│   ├── cache.py                     # Translation cache backends (SQLite/Memory/Null)
│   ├── glossary.py                  # YAML glossary loader and hashing
│   ├── pricing.py                   # Model pricing table + token/cost estimation
│   ├── chart_handler.py             # Chart text collection and update
│   ├── progress.py                  # Rich-based progress display
│   ├── language_detection.py        # 1-shot source language detection (LLM)
│   ├── post_processing.py           # Post-processing utilities
│   ├── config.py                    # Configuration management
│   ├── dependencies.py              # Dependency management
│   ├── text_utils.py                # Text processing utilities
│   └── prompts.py                   # Translation prompts
├── glossary.yaml                    # Default glossary (per-language term map)
├── requirements.txt                 # Python dependencies
├── pyproject.toml                   # Project configuration (uv)
├── uv.lock                          # Dependency lock file
├── .env                             # Environment variables template
├── Dockerfile                       # Docker configuration
├── docs/                            # Documentation
├── imgs/                            # Example images and screenshots
└── samples/                         # Sample files

License

This project is licensed under the MIT License - see the LICENSE file for details.