Speechmatics Server

An MCP (Model Context Protocol) server that gives Claude Code the ability to transcribe audio and video files using the Speechmatics Batch API.

What This Does

Once installed, Claude Code gains access to transcription tools that allow you to:

• Transcribe single files - Convert any audio/video file to text
• Batch transcribe directories - Process entire folders of media files in parallel
• Speaker diarization - Identify different speakers (S1, S2, etc.) in conversations
• Search transcripts - Use Claude's native Grep tool to search across all your transcripts

Example usage in Claude Code:

"Transcribe the meeting recording at ~/Downloads/meeting.mp4"
"Transcribe all the podcasts in ~/Podcasts with speaker identification"
"Search my transcripts for mentions of 'quarterly budget'"

Requirements

• Python 3.11+
• ffmpeg (for audio duration detection)
• Speechmatics API key (free tier available)

Installation

1. Install ffmpeg

# macOS
brew install ffmpeg

# Ubuntu/Debian
sudo apt install ffmpeg

# Windows
winget install ffmpeg

2. Clone and install dependencies

git clone https://github.com/ArchieMcM234/speechmatics_claude_code_mcp.git
cd speechmatics_claude_code_mcp
uv sync

3. Register the MCP server

Add to your Claude Code config file (~/.claude.json):

{
  "mcpServers": {
    "transcription": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/speechmatics_claude_code_mcp",
        "run",
        "python",
        "server.py"
      ],
      "env": {
        "SPEECHMATICS_API_KEY": "your-api-key-here"
      }
    }
  }
}

Replace /path/to/speechmatics_claude_code_mcp with the actual path where you cloned the repo.

Note: Setting the API key in the env block is all you need. You don't need to export it separately or create a .env file.

4. Restart Claude Code

The transcription tools will now be available.

Available Tools

transcribe_file

Transcribe a single audio/video file.

Parameter	Type	Default	Description
file_path	string	required	Absolute path to the media file
accuracy	string	"standard"	"standard" or "enhanced" (enhanced costs more but is more accurate)
diarize	boolean	false	Enable speaker diarization to identify different speakers
with_timestamps	boolean	false	Include word-level timestamps (outputs JSON instead of TXT)
force	boolean	false	Re-transcribe even if a transcript already exists

transcribe_directory

Transcribe all media files in a directory with parallel processing.

Parameter	Type	Default	Description
directory	string	required	Path to directory containing media files
file_types	array	["mp3", "mp4", "wav", ...]	File extensions to include
accuracy	string	"standard"	"standard" or "enhanced"
diarize	boolean	false	Enable speaker diarization
with_timestamps	boolean	false	Include word-level timestamps
force	boolean	false	Re-transcribe even if transcripts exist
recursive	boolean	false	Search subdirectories
max_concurrent	integer	10	Maximum parallel transcription jobs (1-50)

get_transcript

Read an existing transcript file.

Parameter	Type	Description
file_path	string	Path to media file OR transcript file

get_usage

Get Speechmatics API usage statistics for the current month. No parameters required.

Output Formats

Transcripts are saved alongside the original media file.

Plain text (default): filename.transcript.txt

# Transcribed: 2024-01-30T14:32:00Z
# Source: meeting.mp4
# Duration: 12:34
# Accuracy: standard
# Diarization: true

S1: Hello everyone, welcome to the meeting.
S2: Thanks for having me.
...

JSON with timestamps: filename.transcript.json

{
  "metadata": {
    "source": "meeting.mp4",
    "transcribed_at": "2024-01-30T14:32:00Z",
    "duration_seconds": 754,
    "accuracy": "standard",
    "diarization": true
  },
  "transcript": "S1: Hello everyone...",
  "words": [
    {"word": "Hello", "start": 0.0, "end": 0.5, "confidence": 0.98}
  ]
}

Searching Transcripts

After transcribing, Claude can use its native Grep tool to search across all transcripts:

"Search all transcripts in ~/meetings for mentions of 'project deadline'"
"Find where we discussed the budget in the Q4 recordings"

License

MIT