Introduction

Deco is an open-source foundation for building AI-native software.

Create MCP servers that expose tools and workflows to AI agents, with optional custom web interfaces, all deployed on Cloudflare’s global edge network.

Who is it for?

• Vibecoders prototyping ideas
• Agentic engineers deploying scalable, secure, and sustainable production systems

Why Deco?

Our goal is simple: empower teams with Generative AI by giving builders the tools to create AI applications that scale beyond the initial demo and into the thousands of users, securely and cost-effectively.

Core capabilities

• Open-source Runtime – Easily compose tools, workflows, and views within a single codebase
• MCP Mesh (Model Context Protocol) – Securely integrate models, data sources, and APIs, with observability and cost control
• Unified TypeScript Stack – Combine backend logic and custom React/Tailwind frontends seamlessly using typed RPC
• Global, Modular Infrastructure – Built on Cloudflare for low-latency, infinitely scalable deployments. Self-host with your Cloudflare API Key
• Visual Workspace – Build agents, connect tools, manage permissions, and orchestrate everything built in code

Core Concepts

Before diving in, let’s quickly define a few core concepts:

Tools

Functions that can be called by agents, workflows, views and other tools to perform actions like API calls, database queries, or computations.

 const createCepSearchTool = (env: Env) =>
  createTool({
    id: "CEP_SEARCH",
    description: "Search Brazilian postal codes",
    inputSchema: z.object({ cep: z.string() }),
    outputSchema: z.object({ city: z.string(), state: z.string() }),
    execute: async ({ inputData }) => {
      // Tool implementation
    },
  }); 

Workflows

Deterministic sequences that orchestrate multiple steps, that frequently are tools, for creating automations.

 const createCustomerWorkflow = (env: Env) =>
  createWorkflow({ id: "REGISTER_CUSTOMER" })
    .then(validateInput)
    .then(saveToDatabase) 
    .then(sendWelcomeEmail)
    .commit(); 

MCP (Model Context Protocol)

The standard that connects AI models to your tools. Your Cloudflare Worker serves as an MCP server that agents can discover and use.

Agents

AI assistants configured with access to your tools. They understand natural language and decide when to call specific tools to help users.

Views

Optional React frontends that provide custom interfaces for your tools and workflows.