flytohub/flyto-core

Flyto2 Core is the open-source execution kernel for automation and AI-agent workflows: 452 registry-backed modules, MCP-native transport, YAML recipes, evidence capture, replay, triggers, queue, versioning, and metering.

What it solves

Flyto2 Core provides a deterministic execution engine for AI agents and workflow automation. It eliminates the fragility of traditional scripts by allowing users to run complex sequences of browser automation, API calls, and data processing without writing extensive glue code, while providing the ability to to resume from a failed step rather than restarting the entire process.

How it works

The system uses a module-based architecture with a registry of 468 pre-built modules across 85 categories. Workflows are defined as YAML recipes that call these modules. The engine tracks every step's input, output, and timing (Execution Trace), allowing for targeted replay from any specific point of failure. It is MCP-native, meaning AI agents can call these reviewed modules as tools through a standardized schema instead of generating arbitrary code.

Who it’s for

It is designed for developers and AI agent builders who need production-ready browser automation, web scraping, and API integration workflows that are debuggable, replayable, and audit-ready.

Highlights

  • Replayable Workflows: Re-execute from a specific failed step using flyto replay --from-step while preserving context.
  • Extensive Module Library: 468 registry-backed modules covering browser automation (Playwright), AI calls, crypto, and DevOps.
  • MCP-Native: Integrates directly with MCP-compatible clients like Claude Code, Cursor, and Windsurf.
  • Deterministic Execution: Uses reviewed modules and schemas to prevent agents from inventing unreviewed code.
  • Evidence Capture: Automatically captures state snapshots and execution traces for every step.