Shipyard

Ship responsibly.

Human-agent collaboration with receipts.

---

Inbox showing tasks needing review with split panel detail view

The Problem

You're managing multiple AI agents (Claude, Cursor, Devin), but there's no workspace where humans and agents collaborate together:

• No verification — Agent says "done" but you have no proof
• No collaboration layer — Humans review in GitHub, agents work in chat logs
• No feedback loop — You approve work, but the agent never sees it

Shipyard is the collaboration workspace for mixed human-agent teams. Agents create tasks with proof. Humans review in real-time. Feedback flows both ways.

Why Shipyard

• Human-agent collaboration — The first workspace designed for mixed teams. Humans and AI agents work together with structured feedback loops.
• Built for mixed teams — Not a human tool with AI bolted on, or an AI tool ignoring humans. Designed for how agents and humans actually work together.
• Receipts, not promises — Screenshots, videos, and test results. Not just chat logs claiming work was done.
• Zero infrastructure — Works completely locally. GitHub optional for remote artifact sharing. No paid services, no servers to maintain.
• Real-time P2P — Multiple agents and reviewers sync via WebRTC. Works offline, no central server required.

Get Started

Shipyard is just an MCP server. One command or a simple JSON config—works with all major AI coding tools.

Prerequisite: Node.js 22+ (download)

Claude Code (Recommended)

Full experience with hooks, skills, and auto-task creation:

# Step 1: Add the marketplace
/plugin marketplace add https://github.com/SchoolAI/shipyard.git

# Step 2: Install the plugin
/plugin install shipyard@schoolai-shipyard

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "shipyard": {
      "command": "npx",
      "args": ["-y", "-p", "@schoolai/shipyard-mcp@latest", "mcp-server-shipyard"]
    }
  }
}

Codex (OpenAI)

Add via CLI:

codex mcp add shipyard -- npx -y -p @schoolai/shipyard-mcp@latest mcp-server-shipyard

Or add to ~/.codex/config.toml:

[mcp_servers.shipyard]
command = "npx"
args = ["-y", "-p", "@schoolai/shipyard-mcp@latest", "mcp-server-shipyard"]

VS Code / GitHub Copilot

code --add-mcp '{"name":"shipyard","command":"npx","args":["-y","-p","@schoolai/shipyard-mcp@latest","mcp-server-shipyard"]}'

See all platforms (Claude Desktop, Windsurf, JetBrains, Zed, etc.)

Claude Desktop

Add to your config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/claude/claude_desktop_config.json

{
  "mcpServers": {
    "shipyard": {
      "command": "npx",
      "args": ["-y", "-p", "@schoolai/shipyard-mcp@latest", "mcp-server-shipyard"]
    }
  }
}

Windsurf

Add to ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "shipyard": {
      "command": "npx",
      "args": ["-y", "-p", "@schoolai/shipyard-mcp@latest", "mcp-server-shipyard"]
    }
  }
}

JetBrains IDEs

1. Settings > Tools > AI Assistant > Model Context Protocol (MCP)
2. Click "Add"
3. Paste JSON config above

Zed

Add to ~/.config/zed/settings.json:

{
  "context_servers": {
    "shipyard": {
      "command": {
        "path": "npx",
        "args": ["-y", "-p", "@schoolai/shipyard-mcp@latest", "mcp-server-shipyard"]
      }
    }
  }
}

Continue.dev

Create .continue/mcpServers/shipyard.yaml:

mcpServers:
  - name: Shipyard
    command: npx
    args: ["-y", "-p", "@schoolai/shipyard-mcp@latest", "mcp-server-shipyard"]

Full installation guide with troubleshooting →

Platform Support

Platform	Experience
Claude Code	Full integration — hooks, skills, MCP tools, auto-task creation
Codex, VS Code, Cursor, Windsurf	MCP tools — manual workflow, full functionality
Other MCP clients	Basic — depends on platform capabilities

See detailed compatibility matrix →

Using Shipyard skills on other platforms

Shipyard includes a skill (skills/shipyard/SKILL.md) that teaches Claude Code how to use the MCP tools effectively. Other platforms can use this same instruction content through their native mechanisms:

• VS Code / Copilot: Copy to .github/skills/shipyard/ (Agent Skills preview)
• Cursor: Create .cursor/rules/shipyard.mdc with skill content
• Windsurf: Create .windsurf/workflows/shipyard-task.md
• JetBrains: Add to .junie/guidelines.md

See skills/shipyard/README.md for platform-specific instructions.

How It Works

┌─────────────────┐     MCP      ┌─────────────────┐
│   AI Agent      │─────────────►│  MCP Server     │
│ (Claude, etc.)  │              │  (localhost)    │
└─────────────────┘              └────────┬────────┘
                                          │ WebSocket
                                          ▼
┌─────────────────┐   WebRTC    ┌─────────────────┐
│ Remote Reviewer │◄───────────►│ Author Browser  │
│    Browser      │   (P2P)     │                 │
└─────────────────┘             └─────────────────┘

1. Agent creates task via MCP tool → Browser opens automatically
2. Reviewers join via shared URL → Real-time P2P sync
3. Add comments, approve, or request changes → Agent sees feedback
4. Agent uploads artifacts (screenshots, videos) → Stored in GitHub
5. Task auto-completes when all deliverables have receipts (screenshots, videos, test results)

Features

• BlockNote editor — Notion-like editing with inline comments and threads
• Kanban board — Drag-drop tasks between Draft, Review, In Progress, Done
• GitHub artifacts — Screenshots, videos, test results stored in your repo
• Offline-first — IndexedDB persistence, works without network
• Multi-agent — Multiple Claude Code/Cursor instances can work on same task

See more screenshots

Kanban Board

Full Task Detail

Data & Privacy

Data	Where It Lives	Control
Task content	Browser (IndexedDB) + P2P sync	You own it
Artifacts (local)	~/.shipyard/artifacts/ served via localhost	You own it
Artifacts (shared)	Optional: GitHub (your repo, orphan branch)	You own it
MCP server	Runs locally	Never leaves your machine
URLs	Encoded snapshots	Shareable, regenerable

No telemetry. No cloud storage. GitHub optional.

Works completely locally. Add GitHub only if you need remote reviewers to access artifacts.

Documentation

Doc	Description
Installation	Platform-specific setup guides
Setup	Development setup, troubleshooting
Architecture	Data model, sync topology, tech choices
Brief	30-second project context

Architecture

Shipyard is a monorepo with multiple components:

Component	Description
MCP Server	12 tools for task creation, artifacts, feedback
Web App	React + BlockNote editor with Kanban board
Schema	Shared Yjs CRDT types and URL encoding
WebRTC Signaling	P2P discovery (Cloudflare Worker)
Hooks	Claude Code integration

Published package: @schoolai/shipyard-mcp - Includes MCP server + hook

Community

• GitHub Discussions — Questions, ideas, show & tell
• GitHub Issues — Bug reports, feature requests

Contributing

We value ideas over implementations. Please start with discussion:

How to Contribute

1. Bug reports — Open an issue
2. Feature ideas — Start a discussion
3. Questions — Ask in discussions

Before Submitting Code

Open an issue describing what you want to change and get maintainer approval first. This helps us:

• Ensure changes align with project direction
• Avoid duplicate efforts
• Provide design guidance upfront

PRs without a linked, approved issue may be closed.

AI Assistance

AI-assisted contributions are welcome. We use AI ourselves. What matters is that you understand what you're submitting and can answer questions about it.

Learn the Codebase

1. Brief — 30-second project context
2. Engineering Standards — Code quality expectations
3. Architecture — How it all fits together

License

FSL-1.1-ALv2 (Functional Source License)

• Free for all non-competing use
• Converts to Apache 2.0 automatically in 2 years

---