GitMCP Extension

import Tabs from '@theme/Tabs'; import TabItem from '@theme/TabItem'; import YouTubeShortEmbed from '@site/src/components/YouTubeShortEmbed'; import CLIExtensionInstructions from '@site/src/components/CLIExtensionInstructions'; import GooseDesktopInstaller from '@site/src/components/GooseDesktopInstaller';

This tutorial covers how to add the Git MCP Server as a goose extension to give LLMs live access to GitHub repos, enabling smart documentation search, code exploration, and accurate project insights.

:::tip Quick Install <Tabs groupId="interface"> <TabItem value="ui" label="goose Desktop" default> Launch the installer </TabItem> <TabItem value="cli" label="goose CLI"> Command

npx -y mcp-remote https://gitmcp.io/docs

Configuration

Example Usage

In this example, goose uses GitMCP to pull real-time documentation from the openai/whisper GitHub repository, explore how the speech-to-text model works, and surface accurate setup instructions and command-line usage—all directly from the source.

1. Open a terminal and start a new goose session:

goose session

goose Prompt

  Can you help me understand how the OpenAI Whisper speech-to-text model works?

  Use the GitHub repository `openai/whisper` to find accurate information.

  Please:
  - Summarize what the model does and its key features
  - Show an example of how to run it from the command line
  - Explain what files or models are needed to get started
  - Include any useful code snippets or setup instructions you find in the repo

  Give me the response in a Markdown guide format.

goose Output

:::note Desktop

I'll help you understand OpenAI Whisper by exploring their GitHub repository. Let me fetch the documentation and relevant information.

─── owner | repo ──────────────────────────

Now let me search for some specific setup and usage information:

─── owner | query ──────── repo ────────────

Let me search for more technical details and examples:

─── owner | query ──────── repo ────────────

Perfect! Now I have all the information I need from the OpenAI Whisper repository. Let me create a comprehensive Markdown guide for you.

OpenAI Whisper Speech-to-Text Model Guide

Overview Whisper is a general-purpose speech recognition model developed by OpenAI. It's a powerful, multitasking neural network that can perform:

• Multilingual speech recognition (99+ languages)
• Speech translation (convert foreign speech to English text)
• Language identification (detect what language is being spoken)
• Voice activity detection (identify when speech is present)

Key Features:

• ✅ Robust Performance: Trained on 680,000 hours of diverse, multilingual audio data
• ✅ Zero-shot Transfer: Works on many languages without fine-tuning
• ✅ Multiple Model Sizes: From tiny (39M params) to large (1.55B params)
• ✅ Easy to Use: Simple command-line interface and Python API
• ✅ Open Source: MIT licensed with freely available model weights

Architecture Whisper uses a Transformer sequence-to-sequence model that processes audio in 30-second chunks. It converts audio to log-Mel spectrograms and uses special tokens to specify tasks (transcribe, translate, language ID, etc.).

Command-Line Usage Basic Transcription

Getting Started Checklist

1. ✅ Install Python 3.8-3.11
2. ✅ Install FFmpeg for your operating system
3. ✅ Install Whisper: pip install -U openai-whisper
4. ✅ Test installation: whisper --help
5. ✅ Try basic transcription: whisper your_audio_file.mp3

First Run

On your first run, Whisper will automatically download the model weights:

• tiny: ~39 MB download
• base: ~142 MB download
• small: ~461 MB download
• medium: ~1.42 GB download
• large: ~2.87 GB download
• turbo: ~1.5 GB download

Models are cached locally, so subsequent runs are much faster.

Tips & Best Practices

🎯 Model Selection

• English-only: Use .en models (base.en, small.en, etc.)
• Fast transcription: Use tiny or base models
• High accuracy: Use medium or large models
• Translation tasks: Use medium or large (NOT turbo)

🎵 Audio Quality

• Clean, clear audio produces better results
• Whisper handles various audio formats (MP3, WAV, FLAC, M4A, etc.)
• Background noise may affect accuracy

🌍 Language Support

• Supports 99+ languages
• Some languages perform better than others
• Check the language performance breakdown in the repository

💾 Resource Management

• Larger models require more VRAM/RAM
• Consider your hardware limitations when choosing models
• Use smaller models for real-time applications

Whisper is released under the MIT License, making it free to use for both personal and commercial projects.

:::