Skip to content

Installation

PopKit can be installed in three ways depending on your setup:

MethodBest forInstall command
Claude Code pluginClaude Code users/plugin marketplace add jrc1883/popkit-ai
PyPI (MCP server)Cursor, Codex CLI, Copilot, any MCP clientpip install popkit-mcp
PyPI (full)Standalone CLI + all utilitiespip install popkit[full]

Step 1: Add the Marketplace (One-time Setup)

Section titled “Step 1: Add the Marketplace (One-time Setup)”
Terminal window
/plugin marketplace add jrc1883/popkit-ai

Install all plugins for full functionality:

Terminal window
# Install all plugins
/plugin install popkit-core@popkit-ai
/plugin install popkit-dev@popkit-ai
/plugin install popkit-ops@popkit-ai
/plugin install popkit-research@popkit-ai

After installation, restart Claude Code and run /popkit-dev:next to get started.


Install in VS Code Install in VS Code Insiders

Requires pip install popkit-mcp first, then click the badge above.


Install the MCP server for use with Cursor, Codex CLI, GitHub Copilot, or any MCP-compatible client:

Terminal window
pip install popkit-mcp
Terminal window
# Start with stdio transport (default for MCP clients)
popkit-mcp --transport stdio
# Start with HTTP transport for debugging
popkit-mcp --transport streamable-http --port 8080
# Specify custom packages directory
popkit-mcp --packages /path/to/popkit/packages

Cursor — Add to ~/.cursor/mcp.json:

{
"mcpServers": {
"popkit": {
"command": "popkit-mcp",
"args": ["--transport", "stdio"]
}
}
}

Codex CLI — Add to ~/.codex/mcp.json:

{
"mcpServers": {
"popkit": {
"command": "popkit-mcp",
"args": ["--transport", "stdio"]
}
}
}

If you have uv installed, you can run the MCP server without installing:

Terminal window
uvx popkit-mcp --transport stdio

Install everything — CLI, shared utilities, and MCP server:

Terminal window
# Full install (CLI + utilities + MCP server)
pip install popkit[full]
# Or install specific components
pip install popkit # CLI + shared utilities
pip install popkit[mcp] # CLI + utilities + MCP server
pip install popkit-shared # Core utilities only
pip install popkit-cli # CLI only

After installing popkit or popkit-cli:

Terminal window
popkit --help

If you’ve cloned the repository for development and want to test local changes:

Terminal window
# Navigate to the repository root
cd /path/to/popkit-ai
# Install plugins from local directories
/plugin install ./packages/popkit-core
/plugin install ./packages/popkit-dev
/plugin install ./packages/popkit-ops
/plugin install ./packages/popkit-research

Restart Claude Code after installing local plugins.

Note: Local installations take precedence over marketplace installations.


PopKit requires Claude Code 2.1.33+ for full feature support. Tested and verified through CC 2.1.80.

FeatureMinimumNotes
Core functionality2.1.0Skill hot-reload, forked contexts
Plugin auto-update control2.1.2FORCE_AUTOUPDATE_PLUGINS env var
Hook context injection2.1.9PreToolUse additionalContext
Plugin SHA pinning2.1.14Pin to specific git commits
Native task management2.1.16Dependency tracking for agents
Agent memory2.1.32Automatic memory recording
Agent Teams2.1.32Native multi-agent collaboration
Full feature support2.1.33TeammateIdle/TaskCompleted hooks, memory scopes
PostCompact hook2.1.76Hook fires after context compaction
MCP elicitation support2.1.76MCP servers can prompt users for input
Elicitation hooks2.1.76Elicitation + ElicitationResult hook events
Worktree sparse paths2.1.76worktree.sparsePaths config for worktrees
/effort slash command2.1.76Set agent effort level at runtime
Opus 4.6 max output2.1.7764k default / 128k extended output tokens
claude plugin validate2.1.77Improved plugin validation checks
Agent resume via SendMessage2.1.77resume replaced by SendMessage tool
/branch slash command2.1.77/fork renamed to /branch
StopFailure hook event2.1.78Hook fires on abnormal agent stop
${CLAUDE_PLUGIN_DATA} var2.1.78Plugin-scoped persistent data directory
Agent frontmatter extensions2.1.78effort, maxTurns, disallowedTools
Multi-directory seed2.1.79CLAUDE_CODE_PLUGIN_SEED_DIR accepts multiple
1M context support2.1.79Opus 4.6 with 1M context window

Recommended: Always use the latest Claude Code version for best compatibility.

Python requirement: Python 3.11+ for PyPI packages.