pluggedin-plugin

The official Claude Code plugin for Plugged.in -- the AI Infrastructure Platform.

What It Does

This plugin connects Claude Code to the Plugged.in platform, giving you access to 1,500+ MCP tools, persistent memory across sessions, a RAG-powered knowledge base, cross-agent clipboard, notifications, and collective best practices -- all through a single plugin installation. Memory sessions start and end automatically, observations are captured in the background, and relevant memories are injected before context compaction so nothing important is lost.

Quick Start

Prerequisites

• Claude Code (latest version recommended)
• Node.js >= 18.0.0 (for npx)
• A Plugged.in account (free)
• An API key from your Plugged.in dashboard (Settings > API Keys)

Installation

Inside Claude Code, open the plugin manager and add the marketplace:

/plugin marketplace add VeriTeknik/pluggedin-plugin

Then install the plugin:

/plugin install pluggedin@VeriTeknik-pluggedin-plugin

Or use the interactive plugin manager (/plugin) to browse and install from the Discover tab.

The plugin uses the Anthropic MCP proxy to connect to the Plugged.in MCP endpoint. The .mcp.json configuration:

{
  "mcpServers": {
    "pluggedin": {
      "command": "npx",
      "args": ["-y", "@anthropic-ai/mcp-proxy", "--endpoint", "https://mcp.plugged.in/mcp"],
      "env": {
        "PLUGGEDIN_API_KEY": "your-api-key"
      }
    }
  }
}

Setup

After installing, run /pluggedin:setup in Claude Code. The plugin uses a browser-based device authorization flow (similar to gh auth login):

1. Claude calls the Plugged.in API to initiate authorization
2. Your browser opens to a verification page
3. Verify the code shown in your terminal matches the one in the browser
4. Select a Hub and click Authorize
5. The API key is automatically saved to .claude/settings.local.json
6. Restart Claude Code, then run /pluggedin:status to verify

No manual key copying needed -- the entire flow is automatic.

Manual Setup (Fallback)

If the device authorization flow fails (e.g., network issues), you can configure manually:

1. Get your API key from plugged.in/settings (API Keys section) -- it starts with pg_in_
2. Add it to your project's .claude/settings.local.json (gitignored):

{
  "env": {
    "PLUGGEDIN_API_KEY": "pg_in_your_key_here"
  }
}

3. Restart Claude Code and verify with /pluggedin:status

Features

MCP Tools (24 tools via the Plugged.in platform)

Memory (7 tools)

Tool	Purpose
pluggedin_memory_session_start	Start a memory session for the current conversation
pluggedin_memory_session_end	End session and trigger Z-report generation
pluggedin_memory_observe	Record an observation (error, preference, decision, etc.)
pluggedin_memory_search	Semantic search across memories (returns summaries)
pluggedin_memory_details	Get full content for specific memories by UUID
pluggedin_memory_search_with_context	Archetype-enhanced search — returns personal memories + collective patterns filtered through Shadow/Sage/Hero/Trickster based on context
pluggedin_memory_individuation	Get individuation score (0-100) with Memory Depth, Learning Velocity, Collective Contribution, Self-Awareness components and tips

Knowledge Base (1 tool)

Tool	Purpose
pluggedin_ask_knowledge_base	RAG query against your uploaded documents

Clipboard (6 tools)

Tool	Purpose
pluggedin_clipboard_set	Set entry by name or index
pluggedin_clipboard_get	Get entries by name, index, or list all
pluggedin_clipboard_delete	Delete entries or clear all
pluggedin_clipboard_list	List all entries with metadata
pluggedin_clipboard_push	Push to stack (auto-increment index)
pluggedin_clipboard_pop	Pop from stack (LIFO)

Documents (5 tools)

Tool	Purpose
pluggedin_create_document	Create and save AI-generated documents
pluggedin_list_documents	List documents with filtering
pluggedin_search_documents	Search documents semantically
pluggedin_get_document	Retrieve a specific document
pluggedin_update_document	Update or append to a document

Notifications (4 tools)

Tool	Purpose
pluggedin_send_notification	Send custom notifications with optional email
pluggedin_list_notifications	List notifications with filters
pluggedin_mark_notification_done	Mark notification as done
pluggedin_delete_notification	Delete a notification

Discovery (1 tool)

Tool	Purpose
pluggedin_discover_tools	Discover all available tools from configured MCP servers

Skills (9 built-in slash commands)

Skill	Command	Description
Setup	/pluggedin:setup	Browser-based device authorization flow (auto-configures API key)
Status	/pluggedin:status	Check connection status, active session, and memory statistics
Memory Workflow	/pluggedin:memory-workflow	Manage the full memory session lifecycle (start, observe, end)
Memory Search	/pluggedin:memory-search	Semantic search across past memories with progressive disclosure
Memory Status	/pluggedin:memory-status	View memory ring counts, decay stages, and session info
Memory Forget	/pluggedin:memory-forget	Delete specific memories (with confirmation safeguards)
Memory Extraction	(automatic)	Smart background capture of error patterns, decisions, and insights
RAG Context	/pluggedin:rag-context	Search your knowledge base for relevant documentation
Platform Tools	/pluggedin:platform-tools	Complete reference catalog of all available MCP tools

Agents (2 built-in)

Agent	Purpose
Memory Curator	Classifies fresh observations into memory ring types: Procedures (repeatable workflows), Practice (successful habits), Long-term (validated insights), and Shocks (critical failures that bypass decay)
Focus Assistant	Manages a working set of 7+/-2 most relevant items for the current task context, inspired by human working memory limits. Updates the set as task context shifts.

Lifecycle Hooks (5 automatic hooks)

The plugin manages memory sessions and Jungian intelligence automatically through Claude Code lifecycle hooks:

Hook	Trigger	What It Does
SessionStart	Claude Code session begins	Starts a new memory session, displays your individuation score and maturity level
PreToolUse	Before any tool execution	Queries archetype-routed collective patterns and injects relevant suggestions (Shadow for errors, Sage for knowledge, Hero for workflows, Trickster after consecutive failures)
PreCompact	Before context compaction (auto or manual)	Searches for memories relevant to the current project and injects them into context so they survive compaction
PostToolUse (Bash)	After any Bash command execution	Records temporal events for synchronicity detection; captures tool results as observations; queries CBP for known error solutions
Stop	Claude Code session ends	Ends the memory session, triggers Z-report generation, and cleans up temporary state

Jungian Intelligence Layer (v3.2.0)

The plugin integrates four cognitive capabilities inspired by Jungian psychology:

• Synchronicity Detection: Discovers temporal patterns across profiles — tool co-occurrences, failure correlations, and emergent workflows
• Dream Processing: Batch consolidation that clusters and merges semantically similar memories during off-peak hours
• Archetype-Driven Behavior: Context-aware pattern delivery through Shadow (error recovery), Sage (best practices), Hero (workflow guidance), and Trickster (creative alternatives after repeated failures)
• Individuation Metrics: Per-profile maturity scoring (0-100) across Memory Depth, Learning Velocity, Collective Contribution, and Self-Awareness

These features work automatically through the lifecycle hooks — no manual configuration needed.

Architecture

Claude Code
    |
    |-- Lifecycle Hooks (session-start, pre-compact, observe, session-end)
    |       |
    |       v
    |   Plugged.in API (https://plugged.in/api/...)
    |
    |-- MCP Proxy (@anthropic-ai/mcp-proxy)
    |       |
    |       v
    |   Plugged.in MCP Endpoint (https://mcp.plugged.in/mcp)
    |       |
    |       v
    |   1,500+ MCP Tools (memory, knowledge base, clipboard, docs, notifications)
    |
    |-- Skills (slash commands for common operations)
    |
    |-- Agents (background memory management)

• The MCP proxy connects to the Plugged.in MCP endpoint, providing access to all platform tools within Claude Code
• Hooks manage the memory session lifecycle automatically -- no manual intervention needed
• Skills provide slash commands for common operations like searching memory, checking status, and querying the knowledge base
• Agents handle background memory management: classifying observations and maintaining the working set

Configuration

Environment Variables

Variable	Required	Default	Description
PLUGGEDIN_API_KEY	Yes	--	Your Plugged.in API key (starts with pg_in_)
PLUGGEDIN_MCP_ENDPOINT	No	https://mcp.plugged.in/mcp	MCP server endpoint URL
PLUGGEDIN_API_BASE_URL	No	https://plugged.in	API base URL (for self-hosted instances)

Self-Hosted Instances

If running a self-hosted Plugged.in instance, configure all three environment variables:

{
  "env": {
    "PLUGGEDIN_API_KEY": "pg_in_your_key_here",
    "PLUGGEDIN_API_BASE_URL": "https://your-instance.example.com",
    "PLUGGEDIN_MCP_ENDPOINT": "https://your-instance.example.com/mcp"
  }
}

Memory System Overview

The plugin implements a human cognition-inspired memory architecture with concentric rings:

• Fresh Memory: Raw observations captured during a session (unclassified)
• Procedures: Repeatable processes, how-to instructions, step-by-step workflows
• Practice: Repeated successful patterns, habits, coding conventions
• Long-term: Validated insights and facts (requires high confidence)
• Shocks: Critical failures and security events (bypass normal decay)

Memories undergo natural decay through stages: full -> compressed -> summary -> essence -> forgotten. Shock memories are exempt from decay.

Search uses progressive disclosure for token efficiency:

1. Layer 1 (Search): Returns lightweight summaries (50-150 tokens each)
2. Layer 2 (Timeline): Adds temporal context
3. Layer 3 (Details): Returns full memory content

Links

• Platform: https://plugged.in
• Documentation: https://docs.plugged.in
• Plugin Repository: https://github.com/VeriTeknik/pluggedin-plugin
• Main App Repository: https://github.com/VeriTeknik/pluggedin-app
• MCP Proxy Repository: https://github.com/VeriTeknik/pluggedin-mcp

License

MIT