[FEATURE] Plugin marketplace should categorize items by type (MCP, Skill, Agent, Hook) with token cost

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Problem

When browsing a plugin marketplace via /plugin, all items are shown in a flat list with no indication of what type each item is. After installing 21 plugins, I had no way to distinguish between:

• MCP Servers — load tool schemas into context every session (heavy, 5K-15K+ tokens each)
• Skills — lightweight text instructions, loaded on-demand (nearly free)
• Agents — specialized subagent definitions
• Hooks — event-driven shell commands

This matters because the token cost difference is massive. After a bulk install, /doctor reported ~66K tokens of MCP tool context — before I even sent a message. That's a significant chunk of the context window consumed by tool definitions I may rarely use.

Current Experience

1. /plugin → browse marketplace → flat list of 60+ items
2. No type labels (MCP vs Skill vs Agent vs Hook)
3. No token cost indication
4. Install something → no feedback about context budget impact
5. End up with bloated context and no clear way to know what's costing what

Proposed Experience

Categorize items when browsing a marketplace:

/plugin browse claude-plugins-official
├── MCP Servers (⚠️ adds to context budget)
│   ├── aws-serverless (~15K tokens)
│   ├── chrome-devtools (~6K tokens)
│   └── ...
├── Skills (lightweight, on-demand)
│   ├── superpowers
│   ├── frontend-reviewer
│   └── ...
├── Agents
│   └── ...
└── Hooks
    └── ...

Key additions:

• Type labels on every item in the browse/install flow
• Estimated token cost for MCP servers (the main offender)
• Category drill-down so users can browse by type rather than scrolling a flat list
• Install summary showing what was added and its context cost (e.g., "Installed aws-serverless: MCP server, ~15K tokens added to context")

Why This Matters

Users who don't understand the MCP vs Skill distinction (which is most users) will install everything that looks useful, not realizing they're burning 60K+ tokens of context on tool definitions. The /doctor output surfaces this after the fact, but the install flow should make it obvious before committing.

Proposed Solution

Proposed Experience

Categorize items when browsing a marketplace:

/plugin browse claude-plugins-official
├── MCP Servers (⚠️ adds to context budget)
│   ├── aws-serverless (~15K tokens)
│   ├── chrome-devtools (~6K tokens)
│   └── ...
├── Skills (lightweight, on-demand)
│   ├── superpowers
│   ├── frontend-reviewer
│   └── ...
├── Agents
│   └── ...
└── Hooks
    └── ...

Key additions:

• Type labels on every item in the browse/install flow
• Estimated token cost for MCP servers (the main offender)
• Category drill-down so users can browse by type rather than scrolling a flat list
• Install summary showing what was added and its context cost (e.g., "Installed aws-serverless: MCP server, ~15K tokens added to context")

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

Current Experience

1. /plugin → browse marketplace → flat list of 60+ items
2. No type labels (MCP vs Skill vs Agent vs Hook)
3. No token cost indication
4. Install something → no feedback about context budget impact
5. End up with bloated context and no clear way to know what's costing what

Additional Context

_No response_