Where to Find OpenWork Documentation: Complete Guide to different-ai/openwork

OpenWork documentation is distributed across three primary locations: the official website at openworklabs.com/docs for user guides and tutorials, the repository README for installation and quick-start instructions, and the /docs folder for architectural deep-dives and deployment specifications.

The different-ai/openwork repository powers an extensible AI workbench that connects desktop environments to remote capabilities via the Multi-Capability Provider (MCP) protocol. Understanding where to find OpenWork documentation ensures you can correctly install the Electron-based client, configure AI agent integrations, and extend the platform with custom skills.

The Three Primary Sources for OpenWork Documentation

Official Website and User Guides

The primary entry point for OpenWork documentation is the official website at openworklabs.com/docs. This resource contains the complete user-facing guide, including installation steps, UI walkthroughs, and API references for the dev branch. You will also find links to the OpenWork Den control-plane documentation and feature overviews for enterprise deployments.

Repository README for Quick Setup

For immediate installation instructions and project overview, consult the [README.md](https://github.com/different-ai/openwork/blob/dev/README.md) file in the repository root. This document provides concise setup instructions, including the pnpm dev command for local development and JSON configurations for connecting AI agents like Claude Code, Codex, and Cursor via MCP.

Internal Architecture Docs (/docs)

The /docs folder within the repository contains in-depth architectural notes, deployment guides, and internal design specifications. These markdown files detail the Memory Bank storage model, Marketplace capabilities, MCP UI profile handling, and extension manifest schemas required for developing custom plugins.

Key Architectural Components and Source References

Desktop Application Entry Point

The desktop client is an Electron-based application that runs the UI and manages local workspace storage. The entry point is located at packages/app/src/main.ts, which initializes the application window and establishes communication with the OpenWork Den via MCP. Local development instructions in the README reference this package when describing how to launch the application with custom profiles.

OpenWork Den Control Plane

The OpenWork Den serves as the control plane for multi-user, organization-wide management, handling model provisioning and skill publishing. Architectural specifications for this component are documented in docs/memory-bank-architecture.md and docs/marketplace-capabilities-architecture.md, which explain the data persistence model and capability publishing pipelines.

MCP Configuration and Profile Management

The Multi-Capability Provider (MCP) protocol enables remote procedure calls from AI agents to OpenWork capabilities. Profile isolation and keychain handling are detailed in docs/mcp-ui-control-profile.md, which explains how the OPENWORK_DEV_PROFILE environment variable creates isolated worktree environments and mock keychains for credential-free testing. For workflow details regarding external agent interactions, see [AGENTS.md](https://github.com/different-ai/openwork/blob/dev/AGENTS.md).

Skills and Extensions System

Modular functionalities are implemented as skills and plugins, stored in the /.warden/skills directory. The docs/extensions-manifest-foundation.md file defines the JSON schema for extension manifests, specifying how TypeScript or Rust-based skills expose functionalities like Google Workspace OAuth or Azure AKS provisioning.

Essential Configuration Examples

Registering the OpenWork MCP Endpoint

To connect an OpenCode agent, add the following configuration to your agent settings:

{
  "mcp": {
    "openwork": {
      "type": "remote",
      "enabled": true,
      "url": "https://api.openworklabs.com/mcp/agent",
      "oauth": {}
    }
  }
}

This JSON structure registers the remote MCP endpoint that agents use to invoke OpenWork capabilities.

Launching Local Development Instances

Start the Electron application in development mode using pnpm:

pnpm dev

For isolated testing with a specific profile and debug ports:

OPENWORK_DEV_PROFILE=my-feature \
OPENWORK_ELECTRON_REMOTE_DEBUG_PORT=0 \
PORT=0 pnpm dev

The OPENWORK_DEV_PROFILE variable ensures your worktree uses an isolated configuration as described in the MCP UI control profile documentation.

Connecting AI Agents via CLI

Install the OpenWork MCP integration for command-line agents like Codex:

codex mcp add openwork --url https://api.openworklabs.com/mcp/agent

This command registers the endpoint directly with the agent's MCP client.

Summary

• OpenWork documentation is hosted across three tiers: the official website for users, the GitHub README for developers, and the /docs folder for architects.
• The desktop application source resides in packages/app/src/main.ts, while the control plane logic is documented in docs/memory-bank-architecture.md and docs/marketplace-capabilities-architecture.md.
• MCP configuration and profile isolation are governed by specifications in docs/mcp-ui-control-profile.md.
• Extension development requires consulting docs/extensions-manifest-foundation.md for manifest schema definitions.
• Local development uses pnpm dev with optional OPENWORK_DEV_PROFILE environment variables for isolated testing.

Frequently Asked Questions

What is the difference between OpenWork Den and the desktop application?

The desktop application is an Electron-based client that runs locally and manages the user interface, while the OpenWork Den is a remote control plane that handles multi-user management, model provisioning, and skill publishing across organizations. The desktop client communicates with the Den via the MCP protocol as documented in docs/memory-bank-architecture.md.

How do I configure OpenWork for multiple development environments?

Use the OPENWORK_DEV_PROFILE environment variable to create isolated worktree profiles, as specified in docs/mcp-ui-control-profile.md. Each profile maintains separate credentials and settings, allowing you to run multiple development instances simultaneously without configuration conflicts.

Where can I find the schema for creating custom OpenWork extensions?

The extension manifest schema is defined in docs/extensions-manifest-foundation.md in the repository's /docs folder. This document outlines the required JSON structure for registering skills and plugins, including examples for TypeScript and Rust implementations stored in the /.warden/skills directory.

Is there API documentation for the MCP endpoints?

Yes, the MCP endpoint specifications and authentication requirements are documented on the official website at openworklabs.com/docs and referenced in the repository's README.md. The docs/mcp-ui-control-profile.md file provides additional details on profile-specific MCP configurations and mock keychain implementations for testing.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}