flytohub/flyto-core

pip install flyto-core  (core engine + CLI + MCP server)
pip install flyto-core[browser] && playwright install chromium (for browser automation)
Optional: pip install flyto-core[api] for the HTTP API (flyto serve)

Usage

• •Add to MCP config: {"mcpServers": {"flyto-core": {"command": "python", "args": ["-m", "core.mcp_server"]}}} or `claude mcp add flyto-core -- python -m core.mcp_server`
• •Run a built-in recipe: flyto recipe competitor-intel --url https://github.com/pricing, or flyto run my-workflow.yaml
• •Replay from a failed step: flyto replay --from-step 8 (only that step re-executes)
• •HTTP API: flyto serve (127.0.0.1:8333) with POST /v1/workflow/run, /v1/execute, GET /v1/modules, and POST /mcp (MCP Streamable HTTP transport)

Key features

• ✓468 registry-backed modules across 85 catalog categories (browser.*, flow.*, api.*, ai.*, http.*, crypto.*, verify.*, etc.) and 41 built-in YAML recipes
• ✓Execution trace, replay from any step, breakpoints, evidence snapshots, data lineage, and timeout guards
• ✓Deterministic verification modules (verification.* with warroom.* aliases): site graph discovery, replay scenario generation, report packs
• ✓Hardened outbound/file access in the 2.26.x line: guarded HTTP clients against SSRF, sandbox path guard for file/data writes
• ✓Four surface types converge on one engine: CLI, MCP server, HTTP API, and Python API

Best for

Teams that want an auditable, replayable workflow/AI-agent execution engine (browser automation, scraping, evidence capture) with full MCP tool exposure.

Caveats

• ⚠Browser automation relies on Playwright (Chromium install required); Python 3.9+
• ⚠Security-sensitive network/filesystem/auth switches are documented in docs/CONFIGURATION.md; SSRF/path guards may need configuration for allowed use
• ⚠Branding note: use "Flyto2" (not legacy shortenings) in public docs per the repo's public naming contract