ringomcp

ringo-mcp is an MCP server that gives LLM agents a telephone: SIP agents configured in a TOML file, driven from any MCP client (Claude Code, Cursor, Pi, …) over stdio. Place and answer calls, send DTMF, transfer, inspect INVITE headers and media quality — natural language in, SIP out.

you: "Call the support line, wait for the IVR menu and press 2"
 └─ MCP client → ringo-mcp ─┐
                            ├─ dial("support", "0800123456")
                            ├─ wait_event → call_established
                            ├─ send_dtmf("2")
                            └─ …all over SIP

Highlights

• Telephony without glue code — one config file maps SIP accounts to named agents; the LLM addresses them by name.
• Multiple UAs — each agent is its own worker process (own SIP user agent, port and registration), so agents can call each other or the outside world. They start lazily, on first use.
• Event-driven, not guesswork — call control is fire-and-forget; outcomes arrive as events (call_incoming, call_established, call_closed) via the blocking wait_event tool, or by polling agent_status.
• Header round-trip — declare custom headers with per-call templates (${uuid}), add them at runtime, and read the headers of received INVITEs back out.
• Headless by default — no audio hardware needed; agents transmit silence, tones or WAV files via play.

Architecture

ringo-mcp is a thin MCP layer over ringo-agent’s process protocol: one worker process per configured agent, speaking a framed stdio protocol. The binary re-invokes itself (ringo-mcp agent) for those workers, so server and workers always share one build. The same one-process-per-UA design powers ringo-flow’s scenario runner.

MCP client ──stdio(JSON-RPC)── ringo-mcp ──framed stdio── agent worker (SIP UA)
                                             ├─ agent worker (SIP UA)
                                             └─ …

Next steps

• Getting started — install, configure, wire into your MCP client, first call.
• Configuration — the config file reference.
• Tool reference — all tools with their JSON payloads.
• Examples — recipes for the common call flows.

The Rust library API is on docs.rs.