Skip to main content

Components Overview

Components are the building blocks of flows. Each step in a flow executes a specific component that implements a specific operation.

Stepflow provides a set of built-in components for common tasks, and the ability to create custom components using Stepflow SDKs or the Model Context Protocol (MCP).

Types of Components​

1. Built-in Components​

Stepflow provides a variety of built-in components that handle common operations:

  • Data Storage: /put_blob, /get_blob for storing and retrieving data managed by the Stepflow runtime
  • AI Integration: /openai, /create_messages demonstrating interaction with AI APIs
  • File Operations: /load_file demonstrating interaction with the local file system
  • Workflow Control: /eval for executing nested workflows

Learn more about built-in components →

2. Stepflow Component Server​

Use existing copmonent servers or create your own custom components using Stepflow SDKs:

  • Python SDK: Build components in Python with full async support
  • TypeScript SDK: Create high-performance Node.js components (coming soon)
  • Any Language: Implement the Stepflow Protocol directly

Learn more about component servers →.

User-Defined Functions (UDFs)

Many of the SDKs also support user-defined functions (UDFs). These are typically impemented as a component provided by the SDK that takes code as an input and executes it. Learn more about user-defined functions →.

3. MCP Tool Components​

Use tools from Model Context Protocol (MCP) servers as components:

  • Access file systems, databases, and APIs through MCP
  • Leverage the growing ecosystem of MCP tools
  • No additional wrapping needed—MCP tools work directly as components

Learn more about MCP tools →.

Component Architecture​

Most components run in separate processes, ensuring security and stability. During development, these processes can be managed by the Stepflow runtime. In production, these component servers can be separately managed and scaled. For instance, the component server applying specific models can run on a dedicated machine keeping the model in memory for better resource usage.

The diagram below shows how the Stepflow runtime interacts with multiple component servers.

Next Steps​