Skip to main content
Glama
Sign In
deenesjakoruzh
PowerPoint Translator

PowerPoint Translator

by daekeun-ml
OverviewSchemaSourceRelated ServersDiscussionsAdmin
Python
Hybrid

PowerPoint Translator using Amazon Bedrock

A PowerPoint translation tool that calls OpenAI and Anthropic models through Amazon Bedrock Mantle while preserving presentation formatting and structure. Use it as a CLI or a FastMCP server.

Current release: v1.1.0

한국어 · Configuration & MCP reference · CLI cheatsheet

Features

  • GPT-5.6 Sol/Terra/Luna/Cyber and Claude Opus 5/Sonnet 5/Haiku 4.5

  • GPT-5.6 Terra as the default model with optional Luna fallback

  • AWS default credential chain, short-term Mantle tokens, and SigV4

  • Formatting, layout, color, language-specific font, notes, and chart preservation

  • Parallel translation for large presentations and multi-file folder batches

  • Structured PowerPoint-to-Markdown export with optional web verification

  • SQLite or in-memory translation cache

  • Source-language detection, custom YAML glossary, and cost dry-run

  • Exponential retry for transient errors and model fallback after 429/503 failures

  • CLI and FastMCP interfaces

Related MCP server: PowerPoint Translator MCP Service

Examples

The translator preserves complex slide layouts:

Claude Code MCP

Quick Start

Requirements

  • Python 3.11 or later

  • AWS account with Amazon Bedrock access

  • AWS credentials available through the default credential chain

  • Model access in the selected AWS Region

Install

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

Configure AWS credentials:

aws configure
export AWS_REGION=us-east-1
# export AWS_PROFILE=default

The translator generates short-term tokens for GPT models and uses SigV4 for Claude models. AWS_BEARER_TOKEN_BEDROCK is an optional long-term API key override.

The built-in defaults are sufficient for normal use. To customize them:

cp .env.example .env

See Configuration for all settings. The README intentionally does not duplicate the complete .env file.

CLI Usage

Translate an Entire Presentation

uv run ppt-translate translate samples/en.pptx -t ko

Presentations larger than 30 slides are split into parallel chunks:

# Defaults: up to 4 slide workers, 30 slides per chunk
uv run ppt-translate translate large-deck.pptx -t ko

# 80 slides -> 20+20+20+20
uv run ppt-translate translate large-deck.pptx -t ko \
  --slide-workers 4 --slides-per-worker 20

Parallel single-presentation translation

Batch Translate a Folder

# Recursive by default
uv run ppt-translate batch-translate samples/ -t ko

# Top-level files only
uv run ppt-translate batch-translate samples/ -t ko --no-recursive

# Output folder and concurrent PowerPoint files
uv run ppt-translate batch-translate samples/ -t ko -o translated_ko/ -w 10

--workers controls concurrent PowerPoint files. --slide-workers controls parallel slide chunks inside each presentation.

Parallel folder translation

Translate Specific Slides

uv run ppt-translate translate-slides samples/en.pptx -s "1,3,5" -t ko
uv run ppt-translate translate-slides samples/en.pptx -s "2-4" -t ko

Export PowerPoint as Markdown

# AI-structured Korean notes with slide references
uv run ppt-translate export-markdown samples/en.pptx -l ko

# Deterministic extraction without model calls
uv run ppt-translate export-markdown samples/en.pptx \
  --mode extract --language source

# Export every presentation and preserve the folder structure
uv run ppt-translate batch-export-markdown samples/ -l ko

# Verify a bounded set of external claims with client-side web search
uv run ppt-translate export-markdown samples/en.pptx -l ko --web-verify

Structured mode summarizes slides in parallel chunks, creates a sourced presentation overview, and preserves tables, chart data, and speaker notes. Each chunk is validated against the requested output language. Invalid or incomplete responses are regenerated and automatically split into smaller chunks when needed; source-language text is never silently substituted into a translated Markdown document. Web verification is opt-in. See the Markdown export reference.

Dry-Run and Cache

# Estimate tokens and cost without calling the model
uv run ppt-translate translate samples/en.pptx -t ko --dry-run

# Cache is enabled by default at ~/.ppt-translator/cache.db
uv run ppt-translate translate samples/en.pptx -t ko

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

Dry-run cost estimate

The following comparison shows a fully cached folder run and a first run that calls the model:

Batch cache comparison

Other Common Options

# Explicit source language skips auto-detection
uv run ppt-translate translate samples/en.pptx --source-language en -t ko

# Use a glossary; ./glossary.yaml is detected automatically
uv run ppt-translate translate samples/en.pptx -t ko -g glossary.yaml

# Leave chart text unchanged
uv run ppt-translate translate samples/en.pptx -t ko --no-charts

# Inspect the first slides
uv run ppt-translate info samples/en.pptx

See the CLI cheatsheet for the complete everyday command reference.

MCP Server

Start the FastMCP server:

uv run mcp_server.py

Then ask the connected assistant naturally:

Translate samples/en.pptx to Korean
Batch-translate samples/ into Japanese, dry-run first
Show me what is in slide 3
Export samples/en.pptx as structured Korean Markdown

Host configuration for Claude Code and Kiro, plus the complete MCP tool list, is in the Configuration & MCP reference.

Documentation

License

This project is licensed under the MIT License. See LICENSE.

Available Tools

11 tools
batch_export_powerpoint_markdownC

Export all PowerPoint files in a folder as Markdown.

Inputs: input_folder (required), mode, workers, model_id, +9 more. Returns: result

batch_translate_powerpointA

Translate all PowerPoint files in a folder (with optional recursive processing).

Inputs: input_folder (required), dry_run, workers, model_id, +12 more. Returns: result

export_powerpoint_markdownB

Export one PowerPoint presentation as structured Markdown.

Inputs: input_file (required), mode, workers, model_id, +7 more. Returns: result

get_slide_infoC

Get information about slides in a PowerPoint presentation.

Inputs: input_file (required). Returns: result

get_slide_previewB

Get a detailed preview of a specific slide's content.

Inputs: input_file (required), slide_number (required). Returns: result

get_translation_helpA

Get help information about using the PowerPoint translator.

Returns: Help text with usage examples

Inputs: no named parameters. Returns: result

list_supported_languagesA

List all supported target languages for translation.

Returns: List of supported language codes and names

Inputs: no named parameters. Returns: result

list_supported_modelsA

List all supported Amazon Bedrock Mantle models for translation.

Returns: List of supported model IDs

Inputs: no named parameters. Returns: result

post_process_powerpointA

Apply post-processing to a PowerPoint presentation to optimize text boxes.

This function enables text wrapping and shrink text on overflow for text boxes that contain text longer than the specified threshold.

Inputs: input_file (required), output_file, enable_autofit, text_threshold. Returns: result

translate_powerpointC

Translate a PowerPoint presentation to the specified language.

Inputs: input_file (required), dry_run, model_id, output_file, +9 more. Returns: result

translate_specific_slidesB

Translate specific slides in a PowerPoint presentation.

Inputs: input_file (required), slide_numbers (required), dry_run, model_id, +10 more. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A3.5/5.0

    Scored across 11 tools

    Disambiguation4/5

    Most tools are clearly distinct: translate vs. export vs. get info vs. list supported options. However, translate_powerpoint and batch_translate_powerpoint could be confused at first glance, though the batch prefix and description clarify the scope.

    Naming Consistency4/5

    Tool names mostly follow a verb_noun pattern (translate_powerpoint, get_slide_info, list_supported_languages). Minor deviations: get_translation_help uses a noun phrase instead of a verb_noun form, and batch_export_powerpoint_markdown is longer but still consistent.

    Tool Count5/5

    11 tools is well-scoped for a PowerPoint translation server. Each tool covers a distinct operation: translation, export, inspection, and configuration lookups, with no obvious redundancy.

    Completeness4/5

    The server covers the core workflow: translate, batch translate, export, inspect slides, and post-process. Minor gaps include no explicit tool for reverting translations or managing output destinations, but the surface is sufficient for the stated purpose.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Resources

    Related MCP Connectors

    Related MCP Servers

    Appeared in Searches