TranscriptionTools MCP Server

An MCP server providing intelligent transcript processing capabilities, featuring natural formatting, contextual repair, and smart summarization powered by Deep Thinking LLMs.

Available MCP Tools

This MCP server exposes four powerful tools for transcript processing:

1. repair_text - Analyzes and repairs transcription errors with greater than 90% confidence
2. get_repair_log - Retrieves detailed analysis logs from previous repairs
3. format_transcript - Transforms timestamped transcripts into naturally formatted text
4. summary_text - Generates intelligent summaries using ACE cognitive methodology

Installation

Installing via Smithery

To install Transcription Tools for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @MushroomFleet/transcriptiontools-mcp --client claude

1. Clone this repository:

git clone https://github.com/mushroomfleet/TranscriptionTools-MCP
cd TranscriptionTools-MCP

1. Install dependencies:

npm install

1. Build the server:

npm run build

1. Configure the MCP server in your MCP settings file:

{
  "mcpServers": {
    "transcription-tools": {
      "command": "node",
      "args": ["/path/to/TranscriptionTools-MCP/build/index.js"],
      "disabled": false,
      "autoApprove": []
    }
  }
}

Using the MCP Tools

Repairing Transcription Errors

<use_mcp_tool>
<server_name>transcription-tools</server_name>
<tool_name>repair_text</tool_name>
<arguments>
{
  "input_text": "We recieve about ten thousand dollars which is defiantly not enough.",
  "is_file_path": false
}
</arguments>
</use_mcp_tool>

Formatting Timestamped Transcripts

<use_mcp_tool>
<server_name>transcription-tools</server_name>
<tool_name>format_transcript</tool_name>
<arguments>
{
  "input_text": "/path/to/timestamped-transcript.txt",
  "is_file_path": true,
  "paragraph_gap": 8,
  "line_gap": 4
}
</arguments>
</use_mcp_tool>

Generating Summaries

<use_mcp_tool>
<server_name>transcription-tools</server_name>
<tool_name>summary_text</tool_name>
<arguments>
{
  "input_text": "Long text to summarize...",
  "is_file_path": false,
  "constraint_type": "words",
  "constraint_value": 100
}
</arguments>
</use_mcp_tool>

Retrieving Repair Logs

<use_mcp_tool>
<server_name>transcription-tools</server_name>
<tool_name>get_repair_log</tool_name>
<arguments>
{
  "session_id": "20241206143022"
}
</arguments>
</use_mcp_tool>

Core Technologies

Natural Formatting

• Removes timestamps while preserving speech patterns
• Applies intelligent spacing based on pause duration
• Respects natural grammar and language flow
• Maintains exact transcribed content

Contextual Repair

• Identifies and corrects likely transcription errors
• Uses semantic context for high-confidence corrections
• Maintains detailed logs of all changes
• 90% confidence threshold for corrections
• No original audio required

Smart Summarization

• Creates concise summaries of processed transcripts
• Supports multiple constraint types:
  • Time-based (speaking duration)
  • Character count
  • Word count
• Preserves key information and context
• Maintains natural speaking rhythm

Project Structure

/
├── .gitignore         # Git ignore file
├── LICENSE            # MIT license file
├── README.md          # This documentation
├── package.json       # Package dependencies and scripts
├── tsconfig.json      # TypeScript configuration
├── build/             # Compiled JavaScript files (generated after build)
│   ├── tools/         # Compiled tool implementations
│   └── utils/         # Compiled utility functions
└── src/               # Source TypeScript files
    ├── index.ts       # MCP server entry point
    ├── tools/         # Tool implementations
    │   ├── formatting.ts
    │   ├── repair.ts
    │   └── summary.ts
    └── utils/         # Utility functions
        ├── file-handler.ts
        └── logger.ts

Configuration

You can customize the server behavior by modifying the source code directly. The key configuration parameters are found in the respective tool implementation files:

// In src/tools/formatting.ts
const paragraph_gap = 8; // seconds
const line_gap = 4;      // seconds

// In src/tools/repair.ts
const confidence_threshold = 90; // percentage

// In src/tools/summary.ts
const default_speaking_pace = 150; // words per minute

License

MIT

See Also

• TranscriptionTools-MCP — Transcript processing
• DeepLucid3D-MCP — Cognitive processing
• UNO-MCP — Narrative enhancement
• gitea-mcp — Gitea integration
• zero-vector-MCP — Procedural generation