Clinical voice analysis MCP server — AVQI, DSI, jitter/shimmer, pronunciation assessment, and more.
io.github.pmarmaroli/vocametrix-mcp MCP Server
The io.github.pmarmaroli/vocametrix-mcp server is an official Model Context Protocol (MCP) server for the Vocametrix voice analysis API. It provides MCP-compatible AI assistants direct access to clinical voice metrics, pronunciation assessment, speech transcription, and AI-powered therapy planning.
🛠️ Key Features
Clinical voice analysis API access
Clinical voice metrics, including AVQI and DSI
Jitter/shimmer measures
Pronunciation assessment support
Speech transcription
AI-powered therapy planning
🚀 Use Cases
Integrate clinical voice metrics into MCP-based assistants
Run pronunciation assessment workflows
Support speech transcription and downstream analysis
Use AI-assisted therapy planning in healthcare contexts
⚡ Developer Benefits
Standardized MCP server for model-context-protocol clients (e.g., Claude Desktop, Cursor, Cline)
Access to voice-analysis and pronunciation-assessment data via llm-tools
Designed for healthcare-api style integration
⚠️ Limitations
Documentation excerpt provided does not specify tool count, authentication, or deployment details
Gives any MCP-compatible AI assistant (Claude Desktop, Cursor, Cline, etc.) direct access to clinical voice metrics, pronunciation assessment, speech transcription, and AI-powered therapy planning.
Quick start
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):
Get an API key at vocametrix.com/registration. MCP analysis uses purchased API credits; the website subscription and trial do not cover API calls.
ChatGPT (OAuth, API credits)
The optional /chatgpt/mcp endpoint links a user's API account through OAuth. The user approves access on the platform's consent page; ChatGPT receives revocable tokens, not the API key. Existing /mcp API-key clients and local stdio clients remain supported.
See ChatGPT setup and release checks for the matching platform changes, database migration, configuration, and tests. This integration must be deployed and tested in ChatGPT before public submission.
Tools
Voice quality (acoustic)
Tool
Description
vocametrix_calculate_avqi
Acoustic Voice Quality Index (AVQI) — overall dysphonia severity
Assess a folder of WAV files. Reads the server's own filesystem, so it is only registered in stdio/local mode (VOCAMETRIX_MCP_LOCAL_FS=1) — not available on the hosted server
vocametrix://recording-guide — Recording protocols for every tool (sustained vowel, connected speech with language-specific reference sentences, glissando, sustained /s/ and /z/)
interpret_voice_assessment — Generate a clinical SLP-style interpretation report from assessment JSON
compare_pre_post_therapy — Quantified pre/post therapy narrative with metric-by-metric comparison
generate_session_report — SOAP-format progress note from pronunciation assessment data
Audio requirements
Format: WAV (16-bit PCM recommended)
Sustained vowel tasks: 3+ seconds of /a/ phonation
Connected speech tasks: 5–30 seconds of read passage
Minimum sampling rate: 16 kHz
How to pass audio to a tool
The audioPath parameter accepts several input types, but which ones are valid depends on how the MCP server is running:
Input
Hosted / remote server
Stdio / local server (npx, Claude Desktop)
https://... blobUrl from vocametrix_upload_audio
✅ recommended
✅
Public https://... URL to a WAV file
✅
✅
Public URL via vocametrix_ingest_url → returned blobUrl
✅ recommended for URL inputs
✅
data:audio/wav;base64,... data URL
✅
✅
Raw base64 string (≥ 512 chars)
✅
✅
Absolute local path (/home/..., C:\...)
❌ rejected
⚠️ requires VOCAMETRIX_MCP_LOCAL_FS=1
For chat clients that attach audio in the conversation (Claude.ai web/mobile, etc.), the LLM cannot pass an absolute path to a hosted server — it must call vocametrix_upload_audio first with the file content base64-encoded, then pass the returned blobUrl as audioPath to any analysis tool. The MCP descriptions guide the LLM toward this workflow automatically.
For stdio/local deployments where the MCP runs on the user's own machine, set VOCAMETRIX_MCP_LOCAL_FS=1 to allow analysis tools to read absolute local paths directly — convenient for batch processing of files already on disk.
Environment variables
Variable
Required
Description
VOCAMETRIX_API_KEY
Yes
Your Vocametrix API key
VOCAMETRIX_MCP_LOCAL_FS
No
Set to 1 to allow analysis tools to read absolute local file paths (stdio/local deployments only). Default off — local paths are rejected with an actionable error so chat clients are pushed toward the vocametrix_upload_audio → blobUrl workflow.
VOCAMETRIX_MCP_ALLOW_PRIVATE_HOSTS
No
Set to 1 to allow fetching audio URLs whose host resolves to a private, loopback or link-local address (a LAN file server, for instance). Default off, including in stdio/local mode: the caller of an MCP tool is an LLM, and an LLM that has read a hostile page can be talked into pointing these tools at your own network.
Development
bash
git clone https://github.com/Vocametrix/vocametrix-mcp.git
cd vocametrix-mcp
npm install
npm run build
npm test# run unit tests
npm run inspector # test with MCP Inspector
MCP Registry
Listed in the official MCP Registry under io.github.pmarmaroli/vocametrix-mcp. Available for one-click installation in MCP-compatible clients (Claude Desktop, Cursor, Zed, Windsurf, and more).