# flyto-core — Replayable Agent Execution Engine

> flyto-core is an engine with traced, replayable steps and MCP-native modules; verified 270★ and includes 32+ recipes to start.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

```bash
pip install flyto-core[browser] && playwright install chromium
flyto recipe competitor-intel --url https://github.com/pricing
flyto replay --from-step 8
```

## Intro

flyto-core is an engine with traced, replayable steps and MCP-native modules; verified 270★ and includes 32+ recipes to start.

**Best for:** Teams who need step-by-step traces, retries-from-step, and repeatable automation (browser, file, queues)

**Works with:** Python 3.x; optional Playwright for browser recipes; README shows CLI + MCP server install options

**Setup time:** 8-18 minutes

### Key facts (verified)

- GitHub: 270 stars · 48 forks · pushed 2026-05-13.
- License: Apache-2.0 · owner avatar + repo URL verified via GitHub API.
- README-backed entrypoint: `flyto recipe competitor-intel --url https://github.com/pricing`.

## Main

- Model automations as explicit steps so you can trace timing, capture artifacts, and replay only the failing segment.

- Prefer recipes for onboarding: run a built-in recipe before authoring custom pipelines, and keep success criteria explicit.

- Persist artifacts as contracts: save screenshots/reports deterministically so runs can be reviewed like CI logs.

### Source-backed notes

- README quickstart shows `pip install flyto-core[browser] && playwright install chromium` then `flyto recipe ...`.
- README shows replaying from a specific step: `flyto replay --from-step 8`.
- Repo description mentions 412 modules and MCP-native execution with triggers/queue/versioning/metering (verified).

### FAQ

- **Is this just a set of scripts?**: No — it’s an engine with tracing and replay, not ad-hoc shell scripts.
- **Do I need Playwright?**: Only for browser recipes; core engine can run non-browser steps too.
- **How do I keep runs reproducible?**: Pin versions, keep inputs explicit, and persist artifacts + traces per run.

## Source & Thanks

> Source: https://github.com/flytohub/flyto-core
> License: Apache-2.0
> GitHub stars: 270 · forks: 48

---

<!-- ZH -->

## Quick Use

```bash
pip install flyto-core[browser] && playwright install chromium
flyto recipe competitor-intel --url https://github.com/pricing
flyto replay --from-step 8
```

## Intro

flyto-core 是带完整追踪与可回放能力的 agent 执行引擎，内建 MCP-native 模块；已验证 270★，仓库介绍含 412 modules，并提供 32+ recipes 作为上手模板。

**Best for:** 需要逐步追踪、从失败步骤回放/重试，以及可复现自动化（浏览器/文件/队列）的团队

**Works with:** Python 3.x；浏览器 recipes 可选 Playwright；README 描述 CLI + MCP server 的安装/用法

**Setup time:** 8-18 minutes

### Key facts (verified)

- GitHub：270 stars · 48 forks；最近更新 2026-05-13。
- 许可证：Apache-2.0；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中可对照的入口命令：`flyto recipe competitor-intel --url https://github.com/pricing`。

## Main

- 把自动化建模为显式 steps，才能做计时追踪、保存产物，并只回放失败段而不是整条流水线重跑。

- 上手优先用 recipes：先跑内置 recipe 再写自定义 pipeline，并把成功标准写清楚。

- 把产物当契约：截图/报告要确定性落盘，让运行像 CI 一样可审查。

### Source-backed notes

- README 的 quickstart 给出：`pip install flyto-core[browser] && playwright install chromium`，然后 `flyto recipe ...`。
- README 展示按步骤回放：`flyto replay --from-step 8`。
- 仓库描述提到 412 modules 与 MCP-native 执行引擎（triggers/queue/versioning/metering；已验证）。

### FAQ

- **只是脚本集合吗？**：不是。它是带追踪与回放的执行引擎，而非零散 shell 脚本。
- **必须装 Playwright 吗？**：仅浏览器 recipes 需要；核心引擎也可运行非浏览器 steps。
- **如何保持可复现？**：锁版本、显式化输入，并为每次运行持久化产物与 traces。

## Source & Thanks

> Source: https://github.com/flytohub/flyto-core
> License: Apache-2.0
> GitHub stars: 270 · forks: 48


---
Source: https://tokrepo.com/en/workflows/flyto-core-replayable-agent-execution-engine
Author: Agent Toolkit