Name: Flyto Core
Author: flytohub

LABSCO SUMMARY

Replay from a failed step is the feature — a shell script makes you run the whole thing again.

Everything else follows from the trace: because each step's result is recorded, resuming at step eight is instant for the first seven and re-runs only what broke. That matters most on the jobs that are slow at one point and cheap everywhere else — a page load, a screenshot, a Web Vitals capture — which is exactly what the browser modules do.

What it is

A local execution engine with an MCP server: YAML recipes composed from registry-backed modules, executed deterministically, with a trace of every step.

What you get

480 registry-backed modules across 88 catalog categories — browser automation, HTTP and API calls, data transforms, files, crypto, verification, triggers and queue modules
Browser work end to end: launch, goto, evaluate, screenshot, viewport changes, Web Vitals capture and close, each step timed
Replay from a chosen step — earlier steps come back instantly from the trace, and only the failing one re-executes
Packaged recipes you can run by name with parameters, plus your own YAML
Evidence capture: screenshots, metrics and a JSON report saved alongside the run
An HTTP surface with the MCP endpoint at POST /mcp, returning an Mcp-Session-Id on initialization

Requirements

Python 3.10 or higher. pip install flyto-core gets the engine, CLI and MCP server; the browser extra plus playwright install chromium adds browser automation. The published package is flyto-core, registered as io.github.flytohub/flyto-core with stdio transport. The MCP HTTP route uses bearer auth.

Setup effort

One command — pip install flyto-core

See installation options

USE CASES

01

Capturing proof, not just data

Screenshot a page at desktop and mobile viewports, take Web Vitals, and save a JSON report alongside them.

browser.screenshotverify.reportwarroom.report

02

Recovering from a flaky step

Replay from the step that failed instead of re-running the whole pipeline.

03

Giving an agent deterministic tools

Let it call registered modules rather than inventing unreviewed code for each job.

meta.modules.list

04

Turning a recurring chore into a recipe

Write the steps once as YAML and run it by name with parameters afterwards.

mcp.recipe

480 tools

Listed in project docs. Your client may show a different list.

agent.autonomous

agent.chain

agent.tool_use

ai.embed

ai.extract

ai.local_ollama.chat

ai.memory

ai.memory.entity

ai.memory.redis

ai.memory.vector

ai.model

ai.tool

ai.vision.analyze

analysis.html.analyze_readability

analysis.html.extract_forms

BEFORE YOU INSTALL

What to know first

Tools you get

480 tools named in the project's documentation.

Requires

Python 3.10+

Where it runs

All 3 documented clients run the server on your machine.

INSTALLATION

Connect your client

Server setup

TERMINAL

pip install flyto-core

MCP Clients

Claude Code, Cursor, Windsurf

JSON

{
  "mcpServers": {
    "flyto-core": {
      "command": "python",
      "args": [
        "-m",
        "core.mcp_server"
      ]
    }
  }
}

Flyto Core