Level 14: Packaging & Distribution¶

Share your workspace as an installable package — let others use your expertise from their AI assistant.

What you'll learn¶

swarmkit publish — bundle workspace for distribution
swarmkit install — install packages from path, tarball, or URL
swarmkit packages — list installed packages
swarmkit mcp-serve — expose workspaces as MCP tools for AI assistants
Package format and metadata

Publish your workspace¶

1. Add package metadata¶

# package.yaml (workspace root)
name: "@yourname/content-reviewer"
version: 1.0.0
description: >
  Multi-agent content review workspace. Three specialists
  (research, writing, security) coordinate to produce
  thorough content reviews.
author: Your Name <you@example.com>
license: MIT
requires:
  runtime: ">=1.3.0"
  providers:
    - openrouter
  env:
    - OPENROUTER_API_KEY
topologies:
  - content-team
  - structured-review
knowledge:
  searchable: true

2. Bundle it¶

swarmkit publish . --output ./dist

Creates dist/yourname-content-reviewer-1.0.0.tar.gz containing: - Workspace YAML files - Topologies, archetypes, skills - Custom MCP server scripts - Package metadata

Excludes: .env, .swarmkit/, __pycache__/, .git/, *.sqlite.

3. Install a package¶

# From a local directory
swarmkit install ./path-to-workspace/

# From a tarball
swarmkit install ./dist/yourname-content-reviewer-1.0.0.tar.gz

# From a URL (GitHub release)
swarmkit install https://github.com/yourname/content-reviewer/releases/download/v1.0.0/content-reviewer-1.0.0.tar.gz

# Upgrade existing installation
swarmkit install ./updated-workspace/ --upgrade

Packages install to ~/.swarmkit/packages/.

4. List installed packages¶

swarmkit packages

Installed packages:
┌──────────────────────────────┬────────────┬─────────────────────┬─────────────────────────────────┐
│ Package                      │ Topologies │ Installed           │ Path                            │
├──────────────────────────────┼────────────┼─────────────────────┼─────────────────────────────────┤
│ @yourname/content-reviewer   │ 2          │ 2026-06-08          │ ~/.swarmkit/packages/yourname.. │
└──────────────────────────────┴────────────┴─────────────────────┴─────────────────────────────────┘

Expose as MCP tools¶

5. SwarmKit as MCP server¶

Make all installed workspaces available as tools for AI assistants:

swarmkit mcp-serve ./my-swarm

This starts an MCP server on stdio. Each topology becomes a callable tool:

Tools available:
  - run_hello(input: str) — Run Hello World topology
  - run_content_team(input: str) — Run Content Team topology
  - search_knowledge(query: str) — Search workspace knowledge
  - list_workspaces() — Show available workspaces

6. Configure in Claude Desktop¶

// ~/.claude/claude_desktop_config.json
{
  "mcpServers": {
    "swarmkit": {
      "command": "swarmkit",
      "args": ["mcp-serve", "/path/to/my-swarm"],
      "env": {
        "OPENROUTER_API_KEY": "sk-or-..."
      }
    }
  }
}

Now Claude Desktop can use your topologies as tools:

You: Review this PR for security issues

Claude: [calls swarmkit.run_structured_review(input="Review PR #42...")]
        [SwarmKit runs 3-agent code review]

Claude: "The security review found 2 critical issues..."

7. Multiple workspaces¶

Expose multiple workspaces at once:

swarmkit mcp-serve ./workspace1 ./workspace2 ./workspace3

Tools are namespaced to avoid conflicts: - run_workspace1_topology1 - run_workspace2_topology1

Your workspace so far¶

my-swarm/
├── package.yaml            # package metadata
├── workspace.yaml
├── dist/                   # published tarball
│   └── yourname-content-reviewer-1.0.0.tar.gz
└── ...                     # everything from previous levels

Next¶

Level 15: Production Example — a complete workspace that uses every feature.