Cerebro MCP v2.1 Open Source

Describe it.
Cerebro builds it.

Tell Claude what you want in everyday language. Cerebro breaks it into tasks, assigns specialist AI agents, and delivers a working product — no coding required.

See It In Action

View on GitHub


npx @synvoya/cerebro-mcp

MCP Tools

Starter Kits

Zero

Config Files Needed

New in v2.1

Zero config.
Just talk.

Three new capabilities that make Cerebro even easier for non-coders.

⚡

Quick Task

Zero Setup

Just describe what you want. No sessions to create, no agents to configure. Cerebro auto-creates the session, picks the right provider, and delivers.

// You say:

"Build me a to-do app"

// Cerebro:

Done. Files saved to ~/apps/todo

📂

Project Reader

No Blind Edits

Before making changes, Cerebro reads your folder structure and file contents. Understands your code first, edits with full context.

✓Scans directory tree

✓Reads relevant files

✓Ignores node_modules, .git

✓Smart filtering by pattern

🖥️

Visible Terminals

Watch It Work

Each agent task opens a real terminal window showing live progress. See what your AI is doing in real time — no hidden magic.

Designer Agent

$ claude --print "Create layout..."

Building hero section...

Adding menu grid...

✓ Done in 12.3s

How It Works

Three steps.
No coding.

From your idea to a working product, all through conversation.

💬

You Describe

Tell Claude what you want in plain English. Upload screenshots, share ideas, describe features — just like talking to a colleague.

🧠

Cerebro Orchestrates

Cerebro breaks your request into tasks, creates specialist AI agents (Designer, Coder, QA), and coordinates their work automatically.

✨

You Get Results

A working product, tested and ready. Cerebro reports back in plain language — no logs, no terminals, just results.

Live Demo

Watch it work

Pick a scenario and watch Cerebro orchestrate AI agents in real time.

Watch Cerebro turn a simple request into a full website

Cerebro Session

Click Play to start the demo

LIVE SYSTEM FLOW

Conversation

Build me a portfolio website with a blog

On it! Creating Designer, Coder, and QA agents for this project...

Your portfolio is ready! Blog supports markdown posts.

Natural Language

Talk, don't code

Describe what you want the way you'd explain it to a person. No programming languages, no configuration files, no terminal commands.

Describe features in plain English
Upload screenshots of bugs or mockups
Ask for changes naturally — "make the header bigger"
Get results explained in everyday language

Designer

Layouts & UI

Active

Coder

HTML, CSS, JS

Active

Testing & Audit

Active

Marketing

Copy & Social

Active

Agent Swarm

Your AI team,
built by talking

Create specialist agents just by describing what you need. They remember your preferences, delegate tasks to each other, and improve over time.

Create agents with natural language — no config files
Agents delegate to each other automatically
Use starter kits: Web App, API Service, Content Site
Add agents from the community marketplace

🧠

Fresh Context

Every task starts clean

🔐

Session Tokens

Resume across conversations

📊

Context Monitor

Auto-handover when needed

Reliability

Never loses
its place

Each task gets a fresh context window — no confusion from old conversations. When you run out of space, session tokens let you continue seamlessly in a new chat.

Fresh context per task — no accumulated confusion
Signed session tokens survive chat handovers
Agent definitions and preferences persist automatically
Context health monitoring with auto-handover

Capabilities

28 tools.
One install.

Everything you need to go from idea to working product, organized into six categories.

📁

5 tools

Session Management

Create, resume, pause, end, and list sessions

⚡

7 tools

Task Execution

Quick task, execute, read project, status, review, build, test

🤖

7 tools

Agent Operations

Create, update, remove, delegate, install agent packs

🔧

3 tools

Worker Config

Configure workers, detect providers, set model

👁

3 tools

Vision Processing

Analyze images, implement from screenshots, compare

🔄

3 tools

Session Handover

Prepare handover, validate tokens, check context health

Changelog

What’s
shipped

Release history. Click any version to see what changed.

Latest v2.1.0 · Current release

✨ New Features

—Quick Task: zero-setup mode — auto-creates session, picks provider, delivers
—Project Reader: reads folder structure + file contents before editing
—Visible Terminals: each agent task opens a real terminal window with live progress
—Smart Model Routing: configure per-agent / per-provider / per-task models via natural language

🔧 Improvements

—Vision tools fully wired: analyze_image, implement_from_image, compare_screenshots
—Build / test / review tools now spawn real CLI workers
—Agent pack installation loads actual agent definitions from starter kits
—Context health tracking with caution / critical thresholds

🔒 Security

—Token secret now persisted to ~/.cerebro/secret.key (mode 0600)
—Handover tokens now survive process restarts

28 tools · 40 tests passing · 0 stubs

v2.0.0 · Major feature release

✨ New Features

—Multi-Provider CLI: Claude Code, OpenAI Codex, and Aider support with intelligent routing
—Visible Terminal Windows: watch your AI work in real Terminal windows, see every command
—Auto-Close Terminal: terminals stay open for review by default, or pass autoCloseTerminal: true
—Smart Model Routing: Sonnet by default, switch with "use Opus"; Codex auto-uses gpt-5.4
—Auto Git Init for Codex: auto-initializes git repo if needed (Codex requires it)

🔧 Improvements

—Provider & model shown in task results — Chat tells you which agent/provider/model completed
—Task description display expanded to 1000 chars
—Correct model display per provider (no more mismatched model names)

v1.2.0 · Quick Task & Project Reader

✨ New Features

—Quick Task: zero-setup execution — just describe what you want
—Project Reader: new read_project tool reads folder structure + contents
—Provider & model returned in delegate_to_agent response

🔧 Improvements

—Terminal no longer blocks on Enter press (non-blocking I/O)
—Auto-open HTML files in browser after task completion
—Dead code cleanup

v1.0.0 · Initial release · Mar 30, 2026

🚀 Launch

—First public release with 24 MCP tools
—Dual protocol architecture (MCP + A2A)
—Agent swarm with conversation-based creation
—Session management with HMAC-SHA256 signed handover tokens
—Fresh context architecture — no memory rot
—3 starter kits: web-app, api-service, content-site
—Built-in agents: Coder, Designer, Marketer, QA

🏗️ Architecture

—SQLite persistence via better-sqlite3
—Agent Cards published at /.well-known/agent-card.json
—Confidence scoring for automatic agent routing
—Agent heartbeat system with real-time presence
—Message threading between agents

View all releases on GitHub

Starter Kits

Pre-built teams.
Ready to go.

Install a team with one command. Cerebro auto-spawns the right agents for your project.

🌐 Web App

Web Application

Complete team for building web apps with frontend, backend, and testing.

Designer UI / Layout

Coder HTML / CSS / JS

QA Testing / Audit

install web-app pack

🔧 API

API Service

Backend API development with testing and deployment automation.

Coder API / Backend

QA Testing

DevOps CI / Deploy

install api-service pack

📝 Content

Content Site

Marketing sites with design, development, and content generation.

Designer Layout / UI

Coder Static / CMS

Marketing Copy / SEO

install content-site pack

Agent Marketplace

Three tiers. One swarm.

Anthropic Skills

Curated open-source templates from Anthropic's official repository. Verified, quality-checked, free.

Official · Apache 2.0

Community Agents

User-shared agents with reviews, tags, and use-case examples. Expand your swarm with what others have built.

Shared · Reviewed

Your Agents

Agents you create by talking. Private by default, optionally shareable with the community. Yours to keep.

Private · Yours

Get Started

Up and running
in seconds

Install

Terminal

 $ npx @synvoya/cerebro-mcp 

Add to Claude Desktop

Add this to your claude_desktop_config.json:

claude_desktop_config.json

{
  "mcpServers": {
    "cerebro": {
      "command": "npx",
      "args": ["@synvoya/cerebro-mcp"]
    }
  }
}

Restart & Start Building

Restart Claude Desktop. You'll see 28 new tools available. Start with something like:

💬 “Build me a landing page for my bakery”

💬 “Create a Designer and Coder agent for my project”

💬 “Here’s a screenshot of a bug, can you fix it?”

Optional: Configure CLI Providers

Cerebro auto-detects installed tools. For best results, install at least one:

Claude Code Recommended

npm i -g @anthropic-ai/claude-code

OpenAI Codex Alternative

npm i -g @openai/codex

Aider Alternative

pip install aider-chat

Under the Hood

Dual protocol
architecture

Built on two Linux Foundation standards for maximum interoperability.

MCP Model Context Protocol

Handles tool communication between Claude and external systems. CLI tools, databases, GitHub, Slack — all connected through one standard.

- 28 registered tools
- CLI worker spawning
- Image processing pipeline
- Session management

A2A Agent-to-Agent Protocol

Manages agent delegation and external interop. Any A2A-compliant agent — Salesforce, LangChain, CrewAI — can join your swarm.

- Agent-to-agent delegation
- External agent interop
- Agent Card publishing
- Task lifecycle management

You (Claude Chat) → Cerebro MCP → AI Agents + CLI Workers → Working Product

Ready to build
without coding?

Install Cerebro MCP and start building with your words. Open source, free forever.


npx @synvoya/cerebro-mcp

GitHub Repository

NPM Package

📖 Architecture 📘 API Reference 📙 Agent Guide 🤝 Contributing

Apache 2.0 License · Open Source · Made by Synvoya · v2.1.0

Describe it.Cerebro builds it.

Zero config. Just talk.

Quick Task

Project Reader

Visible Terminals

Three steps. No coding.

You Describe

Cerebro Orchestrates

You Get Results

Watch it work

Talk, don't code

Your AI team,built by talking

Never losesits place

28 tools. One install.

Session Management

Task Execution

Agent Operations

Worker Config

Vision Processing

Session Handover

What’s shipped

Pre-built teams. Ready to go.

Web Application

API Service

Content Site

Three tiers. One swarm.

Anthropic Skills

Community Agents

Your Agents

Up and running in seconds

Install

Add to Claude Desktop

Restart & Start Building

Optional: Configure CLI Providers

Dual protocol architecture

Ready to build without coding?

Describe it.
Cerebro builds it.

Zero config.
Just talk.

Three steps.
No coding.

Your AI team,
built by talking

Never loses
its place

28 tools.
One install.

What’s
shipped

Pre-built teams.
Ready to go.

Up and running
in seconds

Dual protocol
architecture

Ready to build
without coding?