Vapi Skills

Agent skills for Vapi — the developer platform for building voice AI agents. These skills follow the Agent Skills specification and can be used with any compatible AI coding assistant including Claude Code, Cursor, VS Code Copilot, Gemini CLI, and more.

Installation

Option 1: npx skills (works with all agents)

npx skills add VapiAI/skills

Install specific skills:

npx skills add VapiAI/skills --skill create-assistant
npx skills add VapiAI/skills --skill create-tool
npx skills add VapiAI/skills --skill create-campaign

Install for a specific agent:

npx skills add VapiAI/skills -a claude-code
npx skills add VapiAI/skills -a cursor

Option 2: Claude Code Plugin (native integration)

/plugin marketplace add VapiAI/skills
/plugin install vapi-voice-ai@vapi-skills

Option 3: Codex Plugin (local development)

The checked-in Codex plugin is assembled from the canonical skill directories, so the Claude and npx skills layouts remain unchanged:

python3 scripts/build-codex-plugin.py
python3 scripts/build-codex-plugin.py --check
python3 scripts/build-codex-plugin.py --archive /tmp/vapi-voice-ai.zip
codex plugin marketplace add .
codex plugin add vapi-voice-ai@vapi-skills

The Codex package is skills-only. The repository-level MCP configuration remains available to agents that support it, but it is not declared as a dependency of this plugin. The assembly step normalizes provider-specific frontmatter and adds OpenAI skill interface settings under each generated skills/<name>/agents/openai.yaml; canonical skill directories remain unchanged. Copy-ready public listing metadata, reviewer tests, asset provenance, release notes, and the remaining account-owner decisions are recorded in codex-plugin/submission-materials.json.

Option 4: Manual installation

Copy any skill directory into your project's .claude/skills/ (for Claude Code), .cursor/skills/ (for Cursor), or the equivalent skills directory for your agent.

Vapi Documentation Server (MCP)

This repository includes configuration for the Vapi documentation MCP server, which gives your AI agent access to the full Vapi knowledge base via RAG. It activates automatically in agents that support MCP.

The skills cover common workflows. The MCP docs server fills in the gaps — advanced configuration, troubleshooting, SDK details, and more.

Supported agents

Agent	Config File	Auto-detected
Claude Code	.mcp.json	Yes
Cursor	.cursor/mcp.json	Yes
VS Code Copilot	.vscode/mcp.json	Yes

Requires: Node.js (for npx). Uses mcp-remote to bridge the remote server. No API key needed for the docs server.

Manual setup

If your agent doesn't auto-detect MCP configs:

Claude Code:

claude mcp add vapi-docs -- npx -y mcp-remote https://docs.vapi.ai/_mcp/server

Any agent (JSON config):

{
  "mcpServers": {
    "vapi-docs": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://docs.vapi.ai/_mcp/server"]
    }
  }
}

Available Skills

Skill	Description
setup-api-key	Guide through obtaining and configuring a Vapi API key
create-assistant	Create voice AI assistants with models, voices, transcribers, tools, and hooks
create-structured-output	Create, attach, run, and verify reusable post-call data extraction
vapi-prompt-builder	Create, improve, or audit production-ready Vapi voice agent and Squad system prompts
create-tool	Build custom tools for assistants — function calls, transfers, integrations
create-call	Initiate outbound phone calls, web calls, and batch calls
create-campaign	Create, schedule, monitor, duplicate, and manage outbound campaigns
create-squad	Build multi-assistant squads with handoff workflows
create-phone-number	Set up phone numbers from Twilio, Vonage, Telnyx, or Vapi
setup-webhook	Configure server URLs to receive real-time call events
simulations	Design, run, and maintain realistic chat and voice simulation suites

Experimental Reference Workflows

Experimental workflows are not included in the standard vapi-voice-ai plugin and are not the recommended path for ordinary Vapi builds. Install one explicitly only when you want to reproduce or adapt that particular architecture.

• vapi-bootstrap-framework — An opt-in recreation of the Bun + TypeScript framework used to build Vapi's landing-page agents from a ROUGH_DRAFT.md

Install it directly with:

npx skills add https://github.com/VapiAI/skills/tree/main/vapi-bootstrap-framework

Configuration

All skills require a Vapi API key. Set it as an environment variable:

export VAPI_API_KEY="your-api-key"

Get your API key from the Vapi Dashboard or use the setup-api-key skill.

SDK Support

API-focused skills pair a conceptual JSON payload with separate implementation examples:

• TypeScript Server SDK — npm install @vapi-ai/server-sdk when the current docs expose the resource method
• Python Server SDK — pip install vapi_server_sdk when the current docs expose the resource method
• TypeScript or Python REST — direct API examples when an SDK resource method is not documented
• cURL — direct REST calls and shell-based verification

The skills do not invent SDK methods. Revalidate language-specific syntax against the current Vapi documentation before producing final implementation code.

Quick Start

1. Get an API key: Use the setup-api-key skill or visit https://dashboard.vapi.ai/org/api-keys
2. Create an assistant: Use the create-assistant skill to build a voice AI agent
3. Add a phone number: Use create-phone-number to get a phone number
4. Make a call: Use create-call to test your assistant

API Reference

• Base URL: https://api.vapi.ai
• Authentication: Bearer token via Authorization: Bearer $VAPI_API_KEY
• Full API Docs: https://docs.vapi.ai
• MCP Docs Server: https://docs.vapi.ai/_mcp/server (auto-configured via .mcp.json)
• API Swagger: https://api.vapi.ai/api

Validation

Validate every skill with the repository's dependency-free validator:

python3 scripts/validate-agent-skills.py
python3 -m unittest discover -s scripts -p 'test_*.py'
python3 scripts/build-codex-plugin.py --check

Pass skill directory names to validate only selected skills. Packaging uses the same validator, so a skill cannot be packaged after failing validation.

License

MIT