Skip to main content

MCP Tools

Model Context Protocol (MCP) provides a standardized way for AI systems to connect to external tools and data sources.

Stepflow's MCP integration allows you to use any MCP-compatible server as a source of components in your workflows, giving you access to file systems, databases, APIs, and other services through a unified interface.

Quick Start​

Installation​

No additional installation is required for MCP support - it's built into Stepflow. However, you'll need the MCP servers you want to use. Many are available as npm packages.

Basic Configuration​

Configure MCP servers as plugins in your stepflow-config.yml:

plugins:
  # MCP filesystem server
  filesystem:
    type: mcp
    command: npx
    args: ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"]

routes:
  "/filesystem/{*component}":
    - plugin: filesystem

Simple Workflow Example​

Use MCP tools in your workflows with the /plugin_name/tool_name format:

schema: https://stepflow.org/schemas/v1/flow.json
input_schema:
  type: object
  properties:
    filename:
      type: string
    content:
      type: string

steps:
- id: write_file
  component: /filesystem/write_file
  input:
    path:
      $from: { workflow: input }
      path: filename
    content:
      $from: { workflow: input }
      path: content

- id: read_file
  component: /filesystem/read_file
  input:
    path:
      $from: { workflow: input }
      path: filename

output:
  file_content:
    $from: { step: read_file }

Next Steps​