# flytohub/flyto-core [Health: Active]

**Category:** ☁️ Cloud Platforms  
**Repository:** https://github.com/flytohub/flyto-core  
**GitHub Stars:** 484  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/flytohub-flyto-core

## Description
Deterministic execution engine for AI agents with 412 modules across 78 categories (browser, file, Docker, data, crypto, scheduling). Features execution trace, evidence snapshots, replay from any step, and supports both STDIO and Streamable HTTP transport.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "flyto-core": {
    "command": "uvx",
    "args": ["flyto-core"]
  }
}
```

## Documentation

## What flytohub/flyto-core does

flytohub/flyto-core is the Python execution engine behind Flyto2. It lets an AI client invoke workflow steps through MCP, while also supporting direct Python use, YAML workflow files, and a local HTTP API. The module inventory covers browser operations, API requests, file handling, data work, tests, and other automation tasks.

The package also includes built-in recipes for repeatable workflows. A recipe can open a website, collect information, capture screenshots, and save results without requiring an agent to generate a new prompt for every run. Workflows can be listed, modules can be browsed, and custom YAML workflows can be executed from the command line.

## How it works

Each workflow run records its individual steps rather than returning only a final result. The resulting execution history can be inspected when a step fails, and checkpoints provide points from which a workflow can be resumed or replayed where appropriate. Evidence reported by a tool can be checked against the action's contract, preserving a record of what the execution produced.

MCP access is available through the package's MCP server entry point. The server supports STDIO and Streamable HTTP transport, allowing it to run locally or be connected to an MCP-compatible AI client. Device-specific integrations remain outside the core execution engine; for example, the ROS 2 setup uses a separate adapter for TurtleBot3.

## Setup and configuration

Install the base package with pip:

```bash
pip install flyto-core
```

For browser workflows, install the browser extra and Chromium:

```bash
pip install "flyto-core[browser]"
playwright install chromium
```

Start the MCP server with:

```bash
python -m core.mcp_server
```

The command-line interface can display available recipes and modules or run a custom workflow:

```bash
flyto recipes
flyto modules
flyto run your-workflow.yaml
```

Python 3.10 or newer is required. No environment variables are needed for the documented setup.

## Tools and capabilities

- MCP access over STDIO or Streamable HTTP
- Browser automation with the browser package extra
- API, file, data, and test-oriented modules
- YAML workflow execution and built-in recipes
- Execution traces and step-by-step run records
- Evidence validation, checkpoints, and replay support
- Local HTTP API and direct Python invocation

## Limitations and notes

Physical-device actions require a trusted host to approve the action. The TurtleBot3 integration uses ROS 2 and Nav2 through a separate adapter; the robot does not run Flyto2 firmware or motor drivers from this repository. The combined physical-robot setup still requires end-to-end hardware sign-off, so the core package should not be treated as a universal plug-and-play robotics layer.

_Full upstream README: https://allmcps.com/mcp/flytohub-flyto-core/readme_