flyto-core is a debuggable automation engine that traces every step of a workflow and allows replay from any point. The server exposes its automation capabilities through a Python package that can be installed via pip and run as a module. Typical usage involves starting the MCP server, then issuing recipes or workflows that leverage the Flyto core’s browser, data, and automation modules. The engine records inputs, outputs, timing, and state at each step, enabling precise reproduction or debugging of complex automation pipelines. You can also use the built-in replay command to re-execute only the steps after a chosen checkpoint, preserving context and reducing re-work when a step fails. The documentation highlights recipes like competitive pricing checks, full audits, and CSV exports, all of which are traceable and replayable.

Prerequisites:

• Python 3.9 or newer
• pip (comes with Python)

Installation steps:

1. Create a virtual environment (recommended) python -m venv venv source venv/bin/activate # On Windows use: venv\Scripts\activate

2. Install flyto-core (core engine + CLI + MCP server) with optional browser support pip install flyto-core

   Optional: include browser automation features

   pip install flyto-core[browser]

3. Install Playwright browsers (for browser automation) pip install playwright playwright install chromium

4. Start the MCP server (from the package module) python -m flyto_core

5. Verify the server is running and accessible via MCP tooling in your environment

Notes and tips:

• The engine provides detailed execution traces, per-step timing, and the ability to replay from any step with preserved context.
• If a step fails (e.g., step 8), you can use flyto replay --from-step 8 to re-run only the failing portion while keeping earlier steps intact.
• When using browser automation, ensure you have the necessary browser dependencies installed (Playwright with Chromium in this guide).
• Configurability: you can set environment variables such as FLYTO_CORE_LOG_LEVEL for logging verbosity and FLYTO_CORE_CONFIG_PATH to point to a custom configuration file.
• Examples of usage are demonstrated in the README via recipes like competitor-intel, full-audit, and scrape-to-csv. These are traceable and replayable end-to-end.
• If you prefer containerized runs, you could wrap the Python module invocation in a docker run command once you have a compatible image built.