sip-phone-mcp

SIP Phone MCP

An MCP server that gives AI agents the ability to make outbound phone calls.

Install

Choose one of three ways:

Option 1: Claude Code plugin (recommended)

claude plugin marketplace add darwincr/sip-phone-mcp
claude plugin install sip-phone-mcp@darwincr-sip-phone-mcp

Updates are automatic. Requires Claude Code v2.1+.

Option 2: Zero-clone with uvx

Add this directly to any MCP client config. No git clone needed -- uvx fetches and runs the package from GitHub automatically:

{
  "mcpServers": {
    "sip-phone": {
      "command": "uvx",
      "args": [
        "--refresh",
        "--from",
        "git+https://github.com/darwincr/sip-phone-mcp.git",
        "sip-phone-mcp"
      ]
    }
  }
}

Remove --refresh if you prefer faster startup over auto-updates.

Option 3: Local install

git clone https://github.com/darwincr/sip-phone-mcp.git && cd sip-phone-mcp
uv tool install -e .

Then point your MCP client at the installed binary:

{
  "mcpServers": {
    "sip-phone": {
      "command": "sip-phone-mcp"
    }
  }
}

Setup

# 1. Copy an example env file and fill in your values
cp .env.example .env          # choose from: .env.example, .env.openai, .env.budget, .env.google, .env.low-latency

# 2. Start the Docker infrastructure
docker compose up --force-recreate -d

Example configs are also included for VS Code (.vscode/mcp.json) and OpenCode (opencode.json).

Preset Env Files

File	Providers	API Keys Needed
.env.example	Deepgram + OpenAI + Cartesia (best quality)	OPENAI, DEEPGRAM, CARTESIA
.env.low-latency	Deepgram + OpenAI + Cartesia (optimized)	OPENAI, DEEPGRAM, CARTESIA
.env.openai	All OpenAI	OPENAI
.env.google	All Google	GOOGLE
.env.budget	Deepgram + OpenAI (cost-effective)	OPENAI, DEEPGRAM

What You Need

• Python 3.11+ and uv
• Docker Desktop
• A SIP trunk provider -- any provider with SIP credentials or sip extension
• API keys for your chosen AI providers (see preset table above)
• Port forwarding is typically not required (outbound calls only). If you experience one-way audio or failed calls, forward these on your router:
  • UDP 5060 (SIP signaling)
  • UDP 30000-30100 (RTP media)

What the AI Can Do

Tool	What it does
make_call	Call a phone number, wait for it to finish, return the transcript and summary
dispatch_voice_agent	Start a call without waiting (for parallel calls)
get_call_status	Check if an async call is still running
get_call_transcript	Get the transcript of a finished call
list_recent_calls	See recent calls and their status
hangup_call	Hang up an active call
get_system_status	Health check for LiveKit connectivity and agent status
get_logs	View recent logs for debugging

Example

"Call +15551234567 and confirm their dental appointment for next Tuesday at 3pm."

The AI uses make_call with instructions for the voice agent, waits for the call to finish, and gets back the full transcript and a summary.

How It Works

AI Client  ──MCP──►  sip-phone-mcp  ──LiveKit API──►  LiveKit + SIP  ──►  Phone Network
                     (MCP server +                     (Docker)
                      voice agent)

Everything runs as a single process. When the MCP server starts, it automatically provisions the SIP trunk, runs smoke tests against your AI providers, and launches the voice agent worker. The Docker stack handles the actual telephony (SIP signaling, media routing).

On each call, the voice agent connects to a LiveKit room, dials out via SIP, and runs a voice conversation loop: STT -> LLM -> TTS. When the call ends, the transcript and an LLM-generated summary are saved.

Logs

~/.sip-phone-mcp/sip-phone-mcp.log

Override the location with the SIP_LOG_DIR env var.