Audit an entire folder of recordings
No external server: the MCP client reads the local folder, sends each file to transcription and audits the risk. You end up with a table to decide.
FalaAI MCPEffort: 2 minutes
Prerequisites
- An MCP client with filesystem access (Claude Code, Cursor, VS Code or Gemini CLI).
- A FalaAI key (fai_...) created in the dashboard.
- A folder with the audio files (mp3, wav, m4a, ogg, opus, webm or aac).
- Credits available: every file is charged.
Configuration
FalaAI · .mcp.json
{
"mcpServers": {
"falaai": {
"type": "http",
"url": "https://mcp-falaai.action.tec.br/mcp",
"headers": { "Authorization": "Bearer fai_..." }
}
}
}Second server
# None. This flow uses only FalaAI MCP: # the folder is read by your own client (local filesystem).
Free test: ask “test the FalaAI health”. If it answers, you are connected.
The request
"Audit every audio file in ./recordings and give me a table with file, risk score and verdict."
What the agent does
List the folderLocal files
The agent lists the files and picks the order (name or date).
Input
Folder path
Output
List of audio files
Transcribe each file · transcribe_audioFalaAI MCP
One file per call: read, base64 and send.
Input
file_base64, filename, language
Output
text and diarized dialog with audio events
Audit the risk · risk_auditFalaAI MCP
The previous step's dialog goes straight into the audit.
Input
dialog, duration_seconds, language; optional response_language
Output
Risk score, verdict, detections and HTML
Build the tableLocal files
The agent merges the results into a table and highlights the worst cases.
Input
Response of each file
Output
Table: file · risk score · verdict
Result
- One row per file, with risk score and verdict.
- Each call's detections stay in the returned JSON.
- The HTML report of each audit comes in the same response.
The folder is not uploaded at once: each file is a call and a charge.
Limitations
- N files = N calls: cost grows with volume.
- The result is not retrievable afterwards: save each file's JSON if you need it.
- No native scheduling: running daily requires an agent or cron on your side.
- Very large folders depend on your MCP client limits.
Variations
- Filter by minimum duration before auditing (avoids spending on short audio).
- Diagnose only (0.03/character) and audit just the negative cases.
- Compare two folders (before and after a training).
Build × buy
Rather not build it? Action implements this flow in your environment, with your account and your access, and validates it with you.