Flyto Core: a workflow engine for agents
Flyto2 Core
A Python engine that runs browser and API work as steps, traces each one and replays from the failed step. It has a CLI and an MCP server with 482 modules
- 484 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Registry modules drive a browser, write and delete files, call external APIs and handle cryptography, and the categories include Docker, Kubernetes and AWS
- A YAML scenario runs as a program: check other people's recipes before running them
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/flyto2-coreDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Registry modules drive a browser, write and delete files, call external APIs and handle cryptography, and the categories include Docker, Kubernetes and AWS
- A YAML scenario runs as a program: check other people's recipes before running them
This is third-party code. Review the repository files before installing.
What it does
Flyto Core runs an agent's work as explicit steps and records what happened in each: input, output, timing and status. If a step fails, flyto replay --from-step 8 repeats only that step and the ones after it. The engine is built as proof: the agent reports a result while the trace and state snapshots show what was done. The registry holds 482 modules in 89 categories: browser control, API calls, data transforms, files, cryptography, branching and loops, verification and screenshot comparison. Ready recipes run with one command, such as flyto recipe full-audit --url https://your-site.com for an SEO, accessibility and speed audit, and the README lists 41 recipes in all. Your own scenarios are written in YAML and run through flyto run. The engine connects to an agent as an MCP server with python -m core.mcp_server and exposes every module as a tool. There is also an HTTP mode, flyto serve, with a POST /mcp route.
Who it is for. For developers and teams who give agents chains of browser and API actions and want an execution trace and replay from the point of failure.
Good fit when
- A chain of browser, API and file writing has to be repeatable and must not restart from scratch when step 8 fails
- You need an execution trace: what went into a step, what came out and how long it took
- You want a quick site audit with the full-audit recipe or competitor price collection with competitor-intel
- A scenario has to run in CI with flyto run workflow.yaml and be shared as a pip package
- You connect a large set of ready modules to Claude Code, Cursor or Windsurf through one MCP server
Not a fit when
- You need one browser command: a dedicated browser control server is quicker
- You are not ready to install Python 3.10+ and Playwright with a Chromium download
- The task is one-off and needs no trace: an orchestration of 482 modules is overkill
- You want to get around site protection: the browser modules are meant for your own and permitted pages
Example request
Run the full-audit recipe for my site and, if a step fails, replay only it and show the traceMCP
- Transport
- stdio, http
- Authentication
- not required
Limitations
- Python 3.10 or newer is required. Browser modules need pip install flyto-core[browser] and playwright install chromium, and the HTTP mode needs flyto-core[api].
- The module and category counts in the README date from publication and change between versions.
- Host-only modules such as capability.invoke cannot be enabled by a YAML parameter: a trusted host grants authority for the run.
- Secrets live in the environment and a credential store rather than in YAML. A template sits in .env.example.
- The license is Apache-2.0.
How to disable. Run claude mcp remove flyto-core and remove the package with pip uninstall flyto-core. Playwright browsers sit in the playwright cache and are removed separately.
FAQ
How do I connect Flyto Core to Claude Code?
Run pip install flyto-core, then claude mcp add flyto-core -- python -m core.mcp_server. For Cursor and Windsurf add a server with the command python and the arguments -m core.mcp_server to the MCP configuration.
What does the flyto replay command do?
It repeats execution from a given step: flyto replay --from-step 8 leaves steps 1-7 untouched and re-runs only the eighth and the following ones. Full context is preserved.
Which ready recipes exist?
The README shows competitor-intel (screenshots and Web Vitals of a pricing page into JSON), full-audit (SEO, accessibility, speed) and scrape-to-csv (selector-based collection into CSV). There are 41 in all.
How does Flyto Core differ from a Playwright script?
A script is rerun in full on failure and gets its logging added by hand. Flyto Core keeps a trace of inputs, outputs and timing, can replay from a step and joins browser, API and files in one description without hand-written glue.
Is there an HTTP interface?
Yes. After pip install flyto-core[api], flyto serve starts a service on 127.0.0.1:8333 with routes for running a workflow, replaying from a step, executing one module, listing modules and POST /mcp.
Discussion
No questions or reviews yet.
Related
MCP serversA self-hosted control plane for running several agents: task dispatch, run review, spend tracking, connected via MCP, CLI or REST
A local-first workflow engine in a single binary with a built-in MCP server so an agent can inspect and run jobs
Zapier's official MCP plugin: the agent gets actions across thousands of apps through your Zapier account
An MCP server for n8n: an assistant creates, runs and deletes workflows through the API, checks executions and tags and requests a security audit
An MCP server for n8n: the agent manages workflows and their executions through the n8n API in natural language, creating, updating, activating and running them
Orchestrator for Claude Code, Codex and Gemini CLI: YAML workflows that loop until your own quality gates pass
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Flyto Core: a workflow engine for agents (this entry) | no | not required | 484 | |
| n8n-MCP | no | key is optional | 23,162 | helps an agent with n8n nodes and templates, while Flyto Core executes steps itself |
| Browser Use | no | 119,424 | an agent drives a browser as it goes, while Flyto Core records steps and replays them |
README in short
The README opens with a promise: AI said it finished, Flyto shows the proof. Then come a 30-second trial with the competitor-intel recipe and step-by-step output, a comparison with Playwright, Selenium and shell scripts, three recipes, installation and an example YAML scenario. Four ways to use it (CLI, MCP, HTTP, Python), a table of 89 module categories and a description of engine features close the document.