standard-vocal-mcp
Enables self-testing of voice agents by running scripted scenarios against an OpenAI-compatible LLM to evaluate behavior, compliance, and quality, returning a score and pass/fail verdict.
standard-vocal-mcp
A Voice Agent Factory exposed over the Model Context Protocol. Not a CRUD wrapper — a factory that deploys production phone agents, tests them, audits their audio, and gates their prompts like code.
Built on top of Vapi (telephony + LLM + STT + TTS). Differs from the official vapi-mcp-server, which is a generic CRUD wrapper over the Vapi API. This is the factory layer: vertical templates, self-testing agents, audio forensics, prompt versioning, CI regression gates.
Why it's hard to copy
The official Vapi MCP is 20 tools that each map to one raw API call. Anyone can read the docs and rebuild it in an hour.
This one encodes months of empirical tuning into its templates — transcriber parameters (numerals, confidenceThreshold, keywords, endpointing), voice selection, conversation flow, urgency handling — plus evaluation logic and audio diagnostics that require knowing the Vapi artifacts API and audio signal processing. The longer it runs, the more reference data it accumulates. That's the moat.
Related MCP server: Clerk Chat MCP Server
Tools
Tool | What it does |
| Lists the assistants on the account (id, name, model, voice, transcriber) — the entry point to find an |
| One tool = one full vertical deployment. |
| The agent tests itself. Fetches the live system prompt, simulates N scripted scenarios against the configured LLM, scores behavior (close speed, no price quoting, spelled digit confirmation, urgency priority, solicitation refusal), returns a |
| Downloads the 3 recording tracks of a call (mono / customer / assistant), runs RMS-per-window analysis, and locates the noise source. Automates the exact manual investigation that found a "computer noise" issue was injected downstream of the caller's mic, not in Vapi. |
| Prompts as code. |
| CI for prompts. Runs |
Quickstart
# Claude Code
claude mcp add standard-vocal -- npx -y standard-vocal-mcp
# Env
export VAPI_TOKEN="your_vapi_token"
export OPENAI_API_KEY="your_openai_key" # used by run_eval to simulate scenarios// claude_desktop_config.json
{
"mcpServers": {
"standard-vocal": {
"command": "npx",
"args": ["-y", "standard-vocal-mcp"],
"env": {
"VAPI_TOKEN": "<token>",
"OPENAI_API_KEY": "<key>"
}
}
}
}Prompt snapshots (prompt_diff) are stored in ./.standard-vocal/ of the directory you launch from — override with STANDARD_VOCAL_STATE_DIR.
Example: one call = one working agent
deploy_agent({ vertical: "elagueur", company: "L'Arbre en Nord", extraKeywords: ["Coutiches", "Ramoniers"] })→ creates the assistant with:
A receptionist prompt tuned for tree service (close after 3 questions, spelled-digit confirmation, urgency priority, never quote a price)
The right French voice (ElevenLabs)
Deepgram
nova-3withnumerals: true,confidenceThreshold: 0.5, and keywords boosted for the business's real streets/towns
Then:
run_eval({ assistantId: "...", vertical: "elagueur" })→ runs 4 scripted scenarios (standard quote, address correction, storm emergency, solicitation) and returns a /100 grade with per-check pass/fail.
Eval report example
GRADE: 100/100 → PASS
[100/100] Devis classique (close rapide)
✓ Demande le type de chantier
✓ Demande l'adresse ou la ville
✓ Confirme le numéro de rue en épelant les chiffres
✓ Ne donne jamais de prix
✓ Clôture avec promesse de rappelArchitecture
Claude / any MCP client
│ (stdio)
▼
standard-vocal-mcp ──┬──► Vapi API (assistants, calls, recordings)
(5 factory tools) ├──► OpenAI-compatible LLM (eval simulation)
└──► Vapi artifacts API (mono/customer/assistant WAV)src/deploy.ts— vertical deploymentsrc/eval.ts— self-testing enginesrc/forensics.ts— audio analysissrc/versioning.ts— prompt history / diff / rollbackverticals/templates.json— vertical templates (prompt, voice, transcriber, keywords)evals/scenarios.json— scripted eval scenarios with weighted checks
Roadmap
More verticals (dentiste, kiné, restaurant, garage)
regression_gateintegration with CI (GitHub Action)Call-level latency breakdown in
audio_forensicsHosted
mcp-remotevariant (like Vapi'smcp.vapi.ai)Scenario auto-generation from real call transcripts
Built by Guillaume Jean Hussong · Standard Vocal — the phone agent factory.
Available Tools
6 toolsaudio_forensicsA
Download the 3 recording tracks of a call (mono/customer/assistant), run RMS analysis, and locate the noise source. Automates the manual audio investigation.
Inputs: callId (required).
deploy_agentA
Deploy a full production phone agent for a market (fr, us, sv) and a vertical (elagueur, plombier, tree-service, plumber, jardineria, plomeria). Creates the Vapi assistant with prompt, voice, transcriber and keywords baked in. One call = one working agent.
Inputs: market (required), company (required), vertical (required), name, +2 more.
list_agentsA
List the Vapi assistants on this account: name, id, model, voice, transcriber, creation date. Start here to find the assistantId for run_eval / prompt_diff / regression_gate.
Inputs: no named parameters.
prompt_diffA
Version and diff assistant prompts. snapshot = capture current prompt from Vapi into local history. diff = compare two versions. rollback = push an old prompt back.
Inputs: action (required), assistantId (required), toVersion, fromVersion.
regression_gateA
CI for prompts. Runs run_eval against the current assistant and blocks the update if the score regressed vs the stored baseline. Returns allow/deny.
Inputs: market (required), vertical (required), assistantId (required), baselineScore (required).
run_evalB
The agent tests itself. Simulates N scripted scenarios via the LLM (no audio cost), scores behavior (close speed, no price, spelled confirmation, urgency handling), returns a /100 report with PASS/WARN/FAIL.
Inputs: market (required), vertical (required), assistantId (required).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 6 tools
Each tool targets a distinct responsibility: listing, deploying, evaluating, analyzing audio, diffing prompts, and gating regressions. Even run_eval and regression_gate are clearly separated, with the former generating scores and the latter comparing against a baseline.
The first three tools follow a verb_noun pattern (list_agents, deploy_agent, run_eval), while the last three are noun-centric (audio_forensics, prompt_diff, regression_gate). This split in grammatical structure makes the naming convention inconsistent and harder to predict.
Six tools is well-scoped for a specialized server focused on voice agent lifecycle and evaluation. Every tool earns its place, and the count is neither too small nor bloated.
The set covers the core workflow: discover agents, deploy new ones, run tests, analyze call audio, version prompts, and gate updates. Minor gaps include no delete_agent or direct update_agent, but these are not critical for the server's apparent purpose.
Maintenance
Related MCP Connectors
AI Voice Production Agent: picks from 30+ TTS models, fixes pronunciation, validates every line.
Build and manage voice AI agents for real phone lines: assistants, calls, numbers and campaigns.
Build, test, deploy and run AI phone agents: agents, numbers, calls, tests, knowledge, campaigns.
Voice agents that book into the software your clients actually run, built and audited from chat.
Related MCP Servers
- AlicenseAqualityDmaintenanceManage voice AI agents, make calls, run campaigns, and control phone numbers through natural language.5827 npm1MIT
- FlicenseAqualityDmaintenanceEnables autonomous prompt improvement for voice AI agents through feedback analysis, test generation, and iterative testing.10-

movoice-mcpofficial
AlicenseAqualityDmaintenanceCreate and manage voice AI agents using natural language from Claude Desktop or Cursor. Enables agent creation, listing, updating, deletion, and viewing call logs.77 npmMIT- AlicenseAqualityBmaintenanceProvides coding agents with visibility into test health through tools for flaky test detection, test quality linting, and LLM evaluation harness, enabling them to triage failures, review test quality, and check prompt changes for regressions.9Apache 2.0