Published Projects & Sites

Static Publishing with ShadowClaw

ShadowClaw combines an interactive browser or Electron client with a headless Node.js agent, a local service layer, and external tool connections. The same agentic core, declarative skills, and tool pipelines can run across each environment.

Documentation

Read the public documentation index, including the architecture overview, CLI and static publishing guide, control plane reference, WebMCP guide, and MCP server reference.

Interactive Client

Run the browser-native PWA or Electron desktop app with orchestration, model calls, and tools isolated in a Web Worker. Client workspaces use OPFS and IndexedDB, with Web Components providing the interface.

# Start the interactive client with live reload
npx shadow-claw dev --open

Headless CLI Agent

Run the same reasoning loop directly from a terminal or CI pipeline. The CLI uses Node.js, native filesystem access, SQLite persistence, host shell execution, and local or cloud model providers.

# Initialize a workspace and run a headless prompt
npx shadow-claw agent init ./my-project
npx shadow-claw agent --workspace ./my-project run "summarize this workspace"

# Inspect available headless-safe tools and skills
npx shadow-claw agent tools
npx shadow-claw agent skills

Server & Control Plane

Run the backend independently of the frontend. The Express service exposes the proxy, scheduler, REST control plane, and MCP endpoint so static sites, browser tabs, CLI commands, and external clients can share a local or remote backend.

# Start services without building or serving a frontend
npx shadow-claw server

Drive Connected Clients

The control plane turns connected browser and Electron clients into remotely addressable execution surfaces. From the CLI, you can list clients, send prompts, inspect scheduled tasks, read client state, and create or manage OPFS workspace backups.

# Discover and operate connected clients
npx shadow-claw clients
npx shadow-claw send "summarize the active workspace" --client <id>
npx shadow-claw tasks --client <id>
npx shadow-claw backup
npx shadow-claw backup list

WebMCP & External Agents

In supported browsers, ShadowClaw registers its tool catalog on document.modelContext through WebMCP and routes execution through the Web Worker. External AI clients can use the official MCP server over Streamable HTTP or the CLI STDIO bridge, while connected clients communicate with the control plane over SSE, WebSocket, or peer-to-peer WebRTC DataChannels.

# Connect an MCP-compatible desktop client over STDIO
npx shadow-claw mcp

# Start a direct peer-to-peer WebRTC listener
npx shadow-claw webrtc listen

Static Publishing

Content repositories use ShadowClaw's dual-root build pipeline to combine the shared runtime bundles with their own pages, assets, configuration, skills, and tools:

# Start local development server on http://127.0.0.1:8888
npx shadow-claw dev

# Or compile static production distribution into ./dist/public
npx shadow-claw build --prod

Deployment Strategies

• Engine repository: xt-ml/shadow-claw is the full ShadowClaw application, CLI, server, and build toolchain. Its own GitHub Pages deployment is the reference app at xt-ml.github.io/shadow-claw.
• Template repository: xt-ml/shadow-claw-template is the reusable starter for content-only sites. It checks out or invokes ShadowClaw in GitHub Actions and publishes the resulting dist/public directory.
• Knowledge hub repositories: block-garden-knowledge-hub and pwgen-knowledge-hub are concrete Strategy B/C content projects. They keep pages, assets, declarative skills, and tools in their own repositories while using ShadowClaw's CLI workflow to deploy interactive knowledge hubs.
• Headless automation and deployment pattern: shadow-claw-agent-cli-weather uses the shadow-claw agent runtime directly for scheduled weather retrieval, markdown report generation, and downstream automation pipelines. It is a deployment-ready pattern for generating structured content for other agents, tooling, or publishing systems.
• Build modes: Any content project can preview with npx shadow-claw dev and publish with npx shadow-claw build --prod, either through the template workflow, a pages-only workflow, or a pinned npm/CLI dependency.