Standalone vs Supercharged Plugin Modes: What's the Difference in Claude Code?

Standalone plugins rely on user input and web search to fulfill requests, while supercharged plugins leverage MCP connectors to automatically pull live data from external services like CRMs and code repositories.

The anthropics/knowledge-work-plugins repository (also referred to as Instagit plugins) provides knowledge work automation for Claude Code and Cowork environments. Understanding the difference between these two operational modes is essential for optimizing your workflow, as each mode determines how your plugins access the data needed to complete tasks.

Core Differences Between Standalone and Supercharged Modes

The fundamental distinction lies in how each mode sources its information. While both modes execute the same skill definitions, their data acquisition strategies create vastly different user experiences.

Data Sources and Connectivity

Standalone mode operates with minimal external dependencies. It processes only the data you explicitly provide and supplements gaps using on-the-fly web searches performed by Claude. No external APIs are invoked, making it ideal for quick tasks where you can paste context directly into the chat.

Supercharged mode transforms the plugin by connecting to external services through MCP (Model Context Protocol) connectors. When enabled, the plugin pulls real-time data from CRMs, source-control platforms, project trackers, calendars, and email systems. For example, in sales/skills/draft-outreach/SKILL.md, the skill description notes that it "Works standalone …, supercharged when you connect …" your CRM, enabling automatic retrieval of prospect histories and recent interactions.

Configuration Requirements

Standalone mode requires zero configuration. The plugin functions immediately in any Claude Code or Cowork environment without setup or credential management.

Supercharged mode requires explicit connector configuration through a .mcp.json manifest file. As seen in partner-built/zoom-plugin/.mcp.json, these manifests define service endpoints, authentication methods (often OAuth2), and required scopes. Users must authorize these connectors before the plugin can access external data, though credentials remain securely stored within the MCP configuration and are never exposed to the plugin code itself.

Performance and User Experience

Standalone mode may require additional back-and-forth conversation to gather missing details, as Claude must prompt you for any information not available through web search.

Supercharged mode delivers faster, richer responses by eliminating manual data entry. When processing a request like "Show me today's stand-up agenda with tickets from JIRA and recent commits from GitHub," the plugin automatically queries connected services rather than asking you to paste the data manually.

How the Architecture Supports Both Modes

The plugin architecture enables this dual-mode functionality through a flexible runtime decision system.

Skill Implementation in SKILL.md Files

Each skill is defined once in a markdown SKILL.md file. According to the source code, these files explicitly document both operational contexts, ensuring users understand when a skill can benefit from external connectors. This single codebase approach means you don't need separate plugins for basic and advanced usage.

Runtime Mode Detection

At execution time, the plugin checks whether required MCP connectors are available in the environment. If the connector is present—as defined in the .mcp.json configuration—the request runs in supercharged mode. If not, the plugin gracefully degrades to standalone mode, falling back to manual user prompts and web search.

Plugin Metadata Structure

The engineering/.claude-plugin/plugin.json file contains plugin metadata used by Claude to load skills and determine available capabilities. This metadata works in conjunction with the MCP manifests to establish which supercharged features are accessible in your specific environment.

Practical Examples: From Standalone to Supercharged

Here is how the same skill behaves differently depending on available connectors.

Standalone Mode Example

When using the draft-outreach skill without connectors:

User: draft outreach to Alice at Acme Corp
Claude: (asks for any specifics you want to include, optionally performs a web search)
Claude: "Here's a personalized email draft based on the information you provided …"

Behind the scenes, the skill reads your prompt, executes a web search for "Alice Acme Corp," and composes the email using only that publicly available data.

Supercharged Mode Example

With a CRM MCP connector enabled:

User: draft outreach to Alice at Acme Corp
Claude (with CRM MCP connected): pulls Alice's record from the CRM, fetches recent interactions, and enriches the email with contextual details automatically.
Claude: "Here's a personalized email draft that references Alice's recent deal and the last meeting …"

The skill calls the connector's API, receives the prospect's full profile, and integrates live data into the generated draft.

Enabling a Connector

To activate supercharged mode, place a .mcp.json file in your plugin directory:

{
  "name": "Zoom",
  "type": "mcp",
  "services": [
    {
      "id": "zoom-meetings",
      "auth": {
        "type": "oauth2",
        "scopes": ["meeting:read", "meeting:write"]
      },
      "endpoint": "https://api.zoom.us/v2"
    }
  ]
}

After authorization, any skill benefiting from Zoom data (such as scheduling meetings) automatically runs in supercharged mode. Reference implementations can be found in partner-built/zoom-plugin/.mcp.json.

Key Files Defining Mode Behavior

Several critical files in the anthropics/knowledge-work-plugins repository govern how standalone and supercharged modes operate:

• engineering/README.md – Explains the standalone/supercharged distinction for engineering workflows and which connectors enhance specific development tasks.

• sales/README.md – Provides sales-specific guidance on when to use standalone mode versus connecting CRM systems for supercharged prospect research.

• sales/skills/draft-outreach/SKILL.md – Demonstrates the standard skill documentation pattern that explicitly states "Works standalone …, supercharged when you connect …" specific external services.

• partner-built/zoom-plugin/.mcp.json – The canonical example of an MCP connector manifest that enables supercharged functionality for video conferencing integrations.

• engineering/.claude-plugin/plugin.json – Defines plugin metadata and capability declarations that Claude uses to determine which mode is available at runtime.

Summary

• Standalone mode operates with user input and web search only, requiring no configuration but potentially needing more manual context sharing.

• Supercharged mode leverages MCP connectors defined in .mcp.json files to pull live data from external services, delivering faster and more contextual results.

• Both modes share identical skill implementations defined in SKILL.md files, with runtime detection determining which data sources are available.

• Security considerations differ between modes: standalone keeps all data within the conversation, while supercharged mode stores OAuth tokens and API keys securely within the MCP connector configuration, isolated from plugin code.

Frequently Asked Questions

Do I need to modify plugin code to switch between standalone and supercharged modes?

No modification is required. The same skill code runs in both modes. The plugin automatically detects whether required MCP connectors are available at runtime. If connectors are present, the skill runs in supercharged mode; otherwise, it gracefully degrades to standalone operation with web search fallbacks.

Where are MCP connector credentials stored?

Credentials are stored securely within the MCP configuration specified in your .mcp.json file. According to the implementation in partner-built/zoom-plugin/.mcp.json, OAuth tokens and API keys are handled by the connector infrastructure and are never exposed to the plugin code itself. This separation ensures that sensitive authentication details remain protected while still enabling data retrieval.

Can I use a plugin in supercharged mode without internet access?

Supercharged mode specifically depends on external API connectivity to services like GitHub, JIRA, or your CRM. While standalone mode can function with limited connectivity (using only provided user input), supercharged mode requires active connections to the external services defined in your MCP connectors. Web searches in standalone mode also require internet access.

How do I know which connectors are available for a specific plugin?

Check the SKILL.md files for individual skills, such as sales/skills/draft-outreach/SKILL.md, which explicitly document the "supercharged when you connect …" requirements. Additionally, the README.md files in both engineering/ and sales/ directories provide comprehensive lists of supported connectors and their specific benefits for different knowledge work workflows.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}