cast-code
v1.0.11
Published
Multi-agent CLI system powered by DeepAgents
Downloads
1,100
Maintainers
Readme
Cast Code
A multi-agent coding CLI for day-to-day engineering work. Cast feels like an AI teammate living inside your terminal — it reads your codebase, plans work, uses tools, and delegates to specialist sub-agents.
Inspired by Claude Code, OpenAI Codex, and Kimi. Built by pedrocastanha.
Why
The main goal is to accelerate product delivery, especially frontend work from design prototypes.
Typical workflow:
- connect Figma Desktop MCP
- map project context with
/init - ask Cast to scaffold screens, components, and styles
- let engineers focus on integration and business logic
Install
npm install -g cast-code
castRequires Node.js >= 20. Works on bash, zsh, fish, dash, ksh and any POSIX-compatible shell on Linux and macOS.
cast not found after install?
The npm global bin directory may not be in your PATH. Find it with npm prefix -g, then add <prefix>/bin to your shell config:
| Shell | File | Line |
|---|---|---|
| bash | ~/.bashrc or ~/.bash_profile | export PATH="$(npm prefix -g)/bin:$PATH" |
| zsh | ~/.zshrc | export PATH="$(npm prefix -g)/bin:$PATH" |
| fish | ~/.config/fish/config.fish | fish_add_path (npm prefix -g)/bin |
| dash / ksh | ~/.profile | export PATH="$(npm prefix -g)/bin:$PATH" |
Reload your shell (source ~/.zshrc, source ~/.bashrc, or open a new terminal), then run cast.
First Run
On first launch the setup wizard runs automatically. To reconfigure at any time:
/config initConfig is stored at ~/.cast/config.yaml.
Providers
Cast supports multiple LLM providers. Configure via /config init or set environment variables:
| Variable | Provider |
|---|---|
| ANTHROPIC_API_KEY | Anthropic / Claude |
| OPENAI_API_KEY | OpenAI / GPT |
| GOOGLE_API_KEY | Google Gemini |
You can also assign different models per purpose — default, subAgent, coder, architect, reviewer, planner, cheap — to tune cost and performance by task type.
Commands
Core
| Command | Description |
|---|---|
| /help | Show command guide |
| /init | Analyze project and generate context |
| /project-deep | Deep context + specialist brief |
| /context | Show session, tools, agents, skills, MCP status |
| /clear | Clear conversation history |
| /compact | Compact context window |
| /exit | Exit CLI |
Config
| Command | Description |
|---|---|
| /config | Config menu |
| /config show | Show current config |
| /config path | Print config file path |
| /config add-provider | Add a new LLM provider |
| /config set-model | Set model by purpose |
MCP
| Command | Description |
|---|---|
| /mcp | MCP hub menu |
| /mcp list | List configured servers |
| /mcp tools | List available tools |
| /mcp add | Add server (from templates or custom) |
| /mcp remove | Remove a server |
| /mcp what | Explain what MCP is |
Git
| Command | Description |
|---|---|
| /status, /diff, /log | Git status, diff, log |
| /commit | AI-generated commit message |
| /up, /split-up | Push / push with split commits |
| /pr | Generate pull request description |
| /review | Code review |
| /fix, /ident, /release | Fix issues, identify patterns, generate release notes |
Agents & Skills
| Command | Description |
|---|---|
| /agents | List specialist agents |
| /agents create | Create a custom agent |
| /skills | List skills |
| /skills create | Create a custom skill |
Mentions
Inject context directly into any prompt:
@src/components/Button.tsx — single file
@src/components/ — entire directory
@git:status — current git status
@git:diff — current diffPlan Mode
For complex requests Cast enters plan mode:
- asks clarifying questions
- generates a structured plan
- lets you refine, approve, or cancel
- executes with the approved plan as context
MCP — Model Context Protocol
Cast ships with templates for 30+ MCP servers across categories: Dev Tools, Design, Data, Search, Cloud, Productivity, Payments, and Browser.
Figma Desktop (recommended)
- Install Figma Desktop
- Open a Design file and enter Dev Mode (
<>button, top right) - In the Inspect panel, enable "Enable desktop MCP server"
- In Cast:
/mcp add→ Design → Figma Desktop - Restart Cast, then
/mcp→ Conectar servidores
For HTTP servers that require OAuth, Cast handles the full OAuth 2.0 + PKCE flow automatically and stores tokens in ~/.cast/mcp-auth/.
Technical Stack
- Runtime: Node.js >= 20, TypeScript
- Framework: NestJS (dependency injection, modular architecture)
- LLM: LangChain + LangGraph (multi-agent orchestration, streaming)
- MCP:
@modelcontextprotocol/sdk(stdio and HTTP/SSE transports, OAuth 2.0 + PKCE) - Providers: Anthropic, OpenAI, Google Gemini, Ollama
- Config: YAML stored at
~/.cast/config.yaml - Auth tokens: stored at
~/.cast/mcp-auth/<server>/
Project layout
src/modules/
repl/ interactive CLI, commands, SmartInput
core/ deep agent orchestration, system prompt, plan mode
agents/ specialist sub-agents (coder, architect, reviewer…)
skills/ skill definitions and knowledge loading
mcp/ MCP client, OAuth provider, server registry, templates
project/ project analysis and context generation
tasks/ task management and plan execution tools
git/ commit, PR, review, and release generators
config/ provider and model configuration
memory/ session memory tools
mentions/ @-mention context injectionLocal Development
npm install
npm run build
npm run startWatch mode:
npm run start:devNotes
- Keep
.cast/context.mdupdated — the richer the context, the better the answers. - Project-level agents and skills live in
.cast/agents/and.cast/skills/at the repo root. - Run
/contextto verify what Cast can currently see.
