Skip to main content
Glama

OpenAct ⚡

The yt-dlp for turning any website or portal into a fast, deterministic REST API, CLI command, and AI Agent (MCP) tool.

License: MIT TypeScript MCP Ready OpenAPI 3.1

Most websites, portals, and SaaS dashboards either don't offer a public API or lock API access behind a $50,000/year "Enterprise Plan." Meanwhile, AI browser agents ("Computer Use") take 45+ seconds, cost $0.50–$2.00 per run, and hallucinate clicks.

OpenAct takes the proven community architecture of yt-dlp and applies it to structured web data and actions:

  1. Declarative .act.yaml Blueprints: A single ~40-line YAML file (with optional TypeScript hooks) defines inputs, extraction logic, output schemas, and automated verification tests.

  2. 3-Tier Execution Engine with Automatic Fallback:

    • Tier 1 (http — 20–80ms): Direct JSON/REST/GraphQL replay with templated inputs and local Session Vault cookies/headers.

    • Tier 2 (extract — 50–150ms): Fast HTML parsing via Next.js/Nuxt Hydration blobs (__NEXT_DATA__), Schema.org JSON-LD (application/ld+json), OpenGraph metadata, and Cheerio CSS selectors—zero headless browser overhead.

    • Tier 3 (browser — Playwright): Full headless Chromium automation and live background XHR/Fetch JSON interception when a site requires JavaScript execution.

  3. 1 Blueprint = 4 Interfaces Automatically:

    • REST API + Live OpenAPI 3.1 Spec (POST /api/v1/run/:namespace/:name)

    • Model Context Protocol (MCP) Server (openact mcp for Claude Desktop, Cursor, Windsurf, OpenHands)

    • Ergonomic CLI (openact run developer/github-trending --input language=typescript)

    • Interactive Web Studio UI (http://localhost:4100)


Built-in Community Blueprints

OpenAct ships with 12 verified blueprints across 4 categories out of the box:

Blueprint ID

Tier

Description

developer/github-trending

extract (CSS)

Trending GitHub repositories by language & time window (daily, weekly, monthly)

developer/github-releases

http + extract fallback

Latest releases for any GitHub repository with automatic HTML fallback

developer/hackernews-search

http

Search Hacker News stories, Ask HN, and Show HN threads

developer/npm-package

http

NPM Registry metadata, version, license, types, and repository URL

developer/pypi-package

http

PyPI Python package metadata, latest version, and Python requirements

knowledge/wikipedia-summary

http

Structured summary, description, and thumbnail for any Wikipedia article

knowledge/dictionary-lookup

http

Wiktionary definitions, parts of speech, and usage examples

knowledge/webpage-metadata

extract (Meta)

Universal OpenGraph, Twitter Card, title, and canonical URL extractor

community/devto-articles

http

Top developer articles and tutorials from DEV.to by tag

community/lobsters-stories

http

Hottest and newest computing stories from Lobste.rs

utilities/open-meteo-weather

http

Real-time weather, temperature, humidity, and wind speed by coordinates

utilities/exchange-rates

http + .hook.ts

Global currency exchange rates with TypeScript transform hook enrichment


Related MCP server: Kryfto

Quick Start

# 1. Install dependencies
npm install

# 2. List all installed community blueprints
npx tsx src/cli/main.ts list

# 3. Run all offline fixture tests (or add --live to test against live websites)
npx tsx src/cli/main.ts test
npx tsx src/cli/main.ts test --live

# 4. Run any blueprint from the CLI
npx tsx src/cli/main.ts run developer/github-trending --input language=typescript --input since=daily

# 5. Launch the OpenAct REST Server, OpenAPI 3.1 Spec & Web Studio UI
npm run serve
# -> Open http://localhost:4100

Reverse-Engineering New Sites in Seconds

OpenAct ships with two built-in compilers so you never have to write a blueprint from scratch:

1. Live URL Inspector (openact inspect <url>)

Automatically probes any URL for direct JSON, __NEXT_DATA__ hydration state, Schema.org JSON-LD, OpenGraph tags, and repeating DOM card/row structures, and outputs a ready-to-run .act.yaml blueprint:

npx tsx src/cli/main.ts inspect https://news.ycombinator.com --save

2. Browser HAR-to-API Compiler (openact import-har <file.har>)

Export a .har file from Chrome/Firefox DevTools ("Save all as HAR with content"), and OpenAct filters out static assets, ranks the underlying JSON API endpoints, infers query inputs and output schemas, and generates a Tier-1 http blueprint:

npx tsx src/cli/main.ts import-har capture.har --save

Local Encrypted Session Vault (AES-256-GCM)

Need to run a blueprint against a portal you're logged into? Store your cookies, headers, or API tokens in OpenAct's local AES-256-GCM encrypted vault (.openact/vault.enc.json) so credentials never touch your YAML files or git history:

# Save headers, cookies, or {{vault.secret}} values locally
npx tsx src/cli/main.ts vault set github \
  --domain api.github.com \
  --header "Authorization=Bearer ghp_xxx"

# List encrypted profiles
npx tsx src/cli/main.ts vault list

Connect to AI Agents (Model Context Protocol)

Every blueprint in blueprints/ is automatically exposed as a deterministic MCP tool:

{
  "mcpServers": {
    "openact": {
      "command": "npx",
      "args": ["tsx", "/path/to/openact/src/cli/main.ts", "mcp"]
    }
  }
}

Contributing Blueprints

OpenAct grows through community-contributed blueprints! Adding support for a new website takes 1 file in blueprints/<category>/<name>.act.yaml. See CONTRIBUTING.md for the 3-minute guide.

License

MIT © OpenAct Contributors

Related MCP Connectors

Related MCP Servers

  • A
    license
    Not graded
    quality
    Not graded
    maintenance
    Exposes over 19,000 Apify Actors as MCP tools for web scraping, data extraction, and OSINT automation. It enables AI agents to dynamically discover and execute scrapers to collect structured data and crawl web content.
    MIT
  • F
    license
    Not graded
    quality
    C
    maintenance
    Provides 42+ MCP tools for browser automation, web scraping, and search, enabling AI agents like Claude and Cursor to browse, extract data, and run research agents on the live web.
    9
    -
  • A
    license
    Not graded
    quality
    B
    maintenance
    Turns the Schema.org markup already present in a webpage into MCP tools an AI agent can call, extracting JSON-LD, microdata, and RDFa to expose read-only and custom tools. It runs as a stdio or Streamable HTTP MCP server, enabling agents to interact with page entities without additional APIs or backends.
    4
    MIT