dazzle

The official CLI and MCP server for Dazzle — cloud stages for AI agents and live streaming.

One binary, two interfaces:

• CLI — full shell access for coding agents (Claude Code, Cursor, terminals) and automation
• MCP server (dazzle mcp) — stdio integration for sandboxed clients (Claude Desktop, VS Code, any MCP host)

Installation

macOS / Linux

curl -sSL https://dazzle.fm/install.sh | sh

Windows (PowerShell)

irm https://dazzle.fm/install.ps1 | iex

Other options

go install github.com/dazzle-labs/cli/cmd/dazzle@latest

Pre-built binaries for macOS (arm64/amd64), Linux (amd64/arm64), and Windows (amd64/arm64) are on the releases page.

Quick Start (CLI)

dazzle login                              # authenticate (opens browser)
dazzle stage create my-stage              # create a stage
dazzle stage up                           # activate — starts streaming
dazzle stage sync ./my-app --watch        # push content, auto-refresh on changes
dazzle stage screenshot -o preview.png    # verify output
dazzle destination add                    # add Twitch/Kick/custom RTMP
dazzle destination attach my-destination  # go live

Quick Start (MCP)

Add to your MCP client config:

{
  "mcpServers": {
    "dazzle": {
      "command": "dazzle",
      "args": ["mcp"]
    }
  }
}

The MCP server starts without credentials — agents can call guide to learn the platform and cli ["login"] to authenticate.

MCP Tools

| Tool | Description | |------|------------| | cli | Run a dazzle CLI command. Use ["--help"] to discover available commands. Output is JSON. | | edit_file | Edit a file in the stage workspace by exact string replacement. The old_string must match exactly once in the file. Use read_file first to see the current content. | | guide | Get the complete Dazzle reference — getting started, CLI commands, content capabilities, and streaming setup. Read this before creating or modifying stage content. | | list_files | List all files in the stage workspace (~/.dazzle/stages/{stage}/). Returns relative paths, one per line. | | read_file | Read a file from the stage workspace (~/.dazzle/stages/{stage}/{path}). | | screenshot | Capture a screenshot of the stage's current browser output. Returns a PNG image. | | sync | Sync the stage workspace (~/.dazzle/stages/{stage}/) to the live stage. Run this after writing files to push content. Equivalent to 'dazzle stage sync {workspace-dir}'. | | write_file | Write a file to the stage workspace (~/.dazzle/stages/{stage}/{path}). Creates parent directories as needed. Use this to build up content that can then be synced to the stage. |

Workspace tools

The workspace tools (write_file, read_file, edit_file, list_files, sync) store files in ~/.dazzle/stages/{stage-id}/ on the host filesystem. This bridges sandboxed environments (e.g. Claude Desktop) where the agent's bash runs in an isolated container and can't share files with the CLI process.

Workflow: write_file → edit_file (iterate) → sync → screenshot (verify)

Limitations: No shell/exec — can't run build tools (npm, tailwind, etc.) in the workspace. Content must be pre-built HTML/CSS/JS. Agents with full filesystem and shell access (e.g. Claude Code) should use dazzle stage sync directly for the full experience.

MCP Resources

| URI | Description | |-----|------------| | https://dazzle.fm/llms-full.txt | Complete Dazzle reference — getting started, CLI help, and content authoring guide. | | https://dazzle.fm/llms.txt | Dazzle quick-start guide — platform overview, setup, CLI basics, and doc links. |

CLI vs MCP — which to use?

| | CLI | MCP | |---|-----|-----| | Best for | Coding agents, terminals, C

... View full README on GitHub