FREE Phone Calls with Claude Code

A hobby VoIP setup can be wired into Claude Code so phone calls—down to an analog payphone—can trigger AI workflows, keep conversational context, and even run business actions like creating ClickUp tasks or generating Slack messages. The core breakthrough is treating SIP (the call-signaling protocol behind most VoIP systems) as the bridge between a phone system and Claude Code, then adding a separate “media” layer for audio and speech-to-text/text-to-speech.

The build starts with three CX’s AI receptionist and transcription features, which the creator sets up in about 10 seconds by adding an API key and configuring an AI agent (“Dolores Umbridge”) with call-handling context. The receptionist can handle both normal requests and edge cases like abusive or frustrated callers, and it supports transcription via OpenAI, Google, or local three CX options. That quick success sparks the bigger idea: if a phone system can talk to Claude Code, then Claude Code can become reachable from anywhere—without requiring the caller to understand anything about APIs.

To make calls work end-to-end, the project leans on SIP for signaling and a separate media server for the actual audio stream. The creator initially considers a commercial all-in-one SIP/media solution (Jam Bones), but balks at the cost—about $1,000 per month for a single node. Instead, the setup uses free, open-source components: FreeSWITCH as the SIP stack and an additional media/processing path built around voice activity detection, Whisper for speech-to-text, and ElevenLabs for text-to-speech. A wrapper server on a Mac handles the handoff: it detects when the user speaks, transcribes, sends the text into Claude Code, then returns Claude Code’s spoken response back through the VoIP pipeline.

A key constraint is that three CX’s free tier doesn’t allow custom SIP trunks. The workaround is to register Claude Code as if it were a phone endpoint inside the three CX system—so it appears as an extension with a ready status and can receive calls directly. From there, the creator defines “call skills” that let Claude Code perform actions during a live call, while Morpheus acts as an executive assistant with access to Claude Code skills.

The practical payoff shows up in demos: Morpheus can create ClickUp tasks and send Slack messages with the task link, while maintaining context across the same call session. It also supports “fire and forget” workflows—calling Morpheus, triggering a job (like generating hyper-realistic thumbnails), then hanging up while results arrive later via Slack.

Finally, the project moves from entertainment to operations with an N8N workflow that checks storage cluster health (e.g., SSD pool capacity thresholds). When conditions are met, an HTTP request triggers the Claude Code-backed phone agent (“Stephanie”) to call the creator, ask for details, and then send a Slack update using its Slack skill. The result is a proof-of-concept for AI-driven phone-based monitoring and response—an interface that can reach the creator even when they’re away from the dashboard.

The creator frames it as a janky but free POC, with a separate documentation/video promised for installation. The bigger message is that once SIP signaling and Claude Code are connected, phone access becomes a control surface for AI—capable of running real workflows, not just answering questions.