Inworld CLI

Authenticate, create API keys, synthesize speech, transcribe audio, and send chat completions from your terminal

Use the Inworld CLI from your terminal to authenticate, create API keys, synthesize speech, generate audiobooks, transcribe audio, call text models through Inworld Inference, and run the Inworld MCP server. Install one global package; no project setup or SDK installation is required.

It's also built for AI coding agents: an agent can log in, mint an API key, and call Inworld APIs directly from the shell — no MCP server configuration needed.

bash

npm install -g @inworld/cli

Requires Node.js 20+. The package installs two commands: inworld (the CLI) and inworld-mcp (the standalone MCP server, equivalent to inworld mcp).

Authentication

bash

# Authenticate with Inworld Platform (opens browser)
inworld login

# Check current authentication status and configuration
inworld status

inworld login opens the Inworld sign-in page, which offers the same methods as the Portal: Google, Microsoft, GitHub, email and SSO. To skip the page, pass --provider google, github or azure (Microsoft) to go straight to that provider, or --provider email to sign in with email and password in the terminal.

Credentials are stored securely on your machine (macOS Keychain, Windows Credential Manager, or an encrypted file on Linux) and refreshed automatically.

CLI 1.4.0 changed how sign-in works. Sign-ins saved by earlier versions no longer work: after upgrading, run inworld login again.

For CI or other non-interactive environments, either authenticate with an Inworld access token or set an API key directly:

bash

# Use an Inworld access token; add --refresh-token "$INWORLD_REFRESH_TOKEN" to let the CLI refresh it
inworld auth login-ci --token "$INWORLD_ACCESS_TOKEN" --email you@example.com

# Or skip login entirely with an API key
export INWORLD_API_KEY="..."

Workspaces and API keys

bash

# Select a workspace for CLI operations
inworld workspace select

# Create a new API key (read-only by default)
inworld workspace add-key --name my-key

# Create a key with read/write permissions for voices and router
inworld workspace add-key --write --name my-key

# Switch the active key to another existing key (add-key makes the new key active)
inworld workspace select-key

# Print the current API key (e.g., to export it as an environment variable)
inworld auth print-api-key

Text-to-Speech

bash

# List available voices
inworld tts list-voices

# Show details for a voice, or listen to its preview sample
inworld tts get-voice Dennis
inworld tts preview Dennis --play

# Synthesize text to speech
inworld tts synthesize "Hello world" --voice Dennis --output hello.mp3

# Synthesize and play through speakers
inworld tts synthesize "[happy] Great to see you!" --play

# Pick a model (inworld-tts-2 or inworld-tts-2-flash)
inworld tts synthesize "Hello world" --model inworld-tts-2-flash --voice Dennis

# Clone a voice from an audio file (3-30 seconds, max 12MB)
inworld tts clone sample.wav --name "My Voice"

# Design a voice from a text description (interactive wizard)
inworld tts design "A warm, deep narrator voice with a slight British accent"

synthesize supports audio markups directly in the text: emotions ([happy], [sad]), delivery styles ([whispering]), vocalizations ([laugh], [sigh]), SSML breaks, and IPA pronunciation.

Audiobooks

Turn a whole book into a single MP3 with chapter markers:

bash

# Generate an audiobook from a .txt or .epub file
inworld audiobook generate book.epub --voice Dennis --output book.mp3

Chapters are detected automatically from Markdown headings or "Chapter 1"-style lines in .txt, or from the spine in .epub. They are synthesized in parallel, each on its own TTS session so voice delivery stays consistent within a chapter. Long runs are resumable: pass --work-dir <dir> and a re-run skips chapters that already finished.

Speech-to-Text

bash

# Transcribe an audio file
inworld stt transcribe recording.wav

# Transcribe with a specific language
inworld stt transcribe recording.mp3 --lang es-ES

Text inference

bash

# List models available to your workspace
inworld llm models

# Send a chat completion request
inworld llm chat "What is the capital of France?" --model gpt-4o-mini

MCP server

The CLI includes the full Inworld MCP server. It gives your AI assistant tools for speech synthesis and TTS previews, voice cloning, design and publishing, STT transcription, LLM chat (with or without spoken audio), workspace search and limits, billing summary, router management, and documentation search.

bash

# Run as a stdio MCP server (for Claude Code, Cursor, etc.)
inworld mcp        # or the standalone bin: inworld-mcp

# Or run over HTTP (listens on 127.0.0.1 by default; the endpoint is POST /mcp)
inworld mcp --http --port 3333

For example, to add it to Claude Code:

bash

claude mcp add inworld -- inworld mcp

The server uses the credentials it finds when it starts, and registers only the tools those credentials can call:

Credential	How to provide it	Tools
API key	`INWORLD_API_KEY`, or the key stored by `inworld workspace select` or `inworld workspace select-key`	Speech synthesis, voice listing, cloning, design and publishing, transcription, chat completions, router listing, and documentation search (`search_docs`)
Sign-in	`inworld login` (or `inworld mcp login`)	Workspace, billing, voice and router management

With both, the server registers every tool. If you add a credential later, restart the MCP server to pick up the new tools.

Voice cloning, design and publishing write to your workspace, so they need a read/write key (inworld workspace add-key --write). With a read-only key the tools are still listed, but those calls fail.

Many agents don't need the MCP server at all — they can call inworld commands directly from the shell.

Telemetry

The CLI collects anonymous usage analytics. To opt out:

bash

inworld auth disable_cli_telemetry

Or set DISABLE_TELEMETRY=1 in the environment.

Inworld CLI vs. Runtime CLI

The Inworld CLI (inworld) covers platform APIs: auth, API keys, TTS, STT, and text inference. For Agent Runtime graph development and deployment (inworld-runtime init/run/deploy), see the Runtime CLI.