Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
112 lines (86 loc) · 2.71 KB

ARCHITECTURE.md

File metadata and controls

112 lines (86 loc) · 2.71 KB

OpenPlugin Architecture

Overview

OpenPlugin is a vendor-agnostic plugin framework that allows you to:

  • Load and manage plugins with standardized structure
  • Execute plugin commands and agents using any LLM provider
  • Integrate with MCP (Model Context Protocol) servers
  • Build extensible AI applications

Core Components

Plugin Manager

The PluginManager is the central component that:

  • Discovers and loads plugins from directories
  • Manages plugin lifecycle
  • Coordinates command/agent execution
  • Handles MCP client connections

Plugin

The Plugin class represents a loaded plugin:

  • Parses plugin manifest (plugin.json)
  • Loads commands, agents, and skills
  • Manages MCP configuration
  • Provides access to plugin resources

Providers

Providers implement the LLMProvider interface:

  • OpenAIProvider: OpenAI GPT models
  • Future: AnthropicProvider, GoogleProvider, etc.
  • Custom providers can be implemented

MCP Support

MCP (Model Context Protocol) integration:

  • MCPClient: Communicates with MCP servers
  • Supports stdio transport
  • Tool discovery and execution
  • Automatic tool conversion for providers

Data Flow

User Input
    ↓
PluginManager.execute_command()
    ↓
Plugin.get_command() → Command Content
    ↓
MCPClient.list_tools() → Available Tools
    ↓
Provider.execute_command() → LLM API Call
    ↓
Response with Tool Calls (if any)
    ↓
MCPClient.call_tool() → Tool Execution
    ↓
Final Response

Plugin Discovery

Plugins are discovered by:

  1. Scanning the plugins directory
  2. Looking for .claude-plugin/plugin.json files
  3. Validating plugin structure
  4. Loading plugin resources

MCP Integration

MCP servers are configured in .mcp.json:

  • Each server has a command and arguments
  • Servers run as subprocesses
  • Communication via JSON-RPC 2.0 over stdio
  • Tools are automatically discovered and made available

Provider Abstraction

The LLMProvider interface abstracts LLM differences:

  • Standardized command/agent execution
  • Tool/function calling support
  • Message format conversion
  • Provider-specific optimizations

Extension Points

Adding a New Provider

  1. Implement LLMProvider interface
  2. Convert MCP tools to provider format
  3. Handle provider-specific features
  4. Register in providers/__init__.py

Adding Plugin Features

  1. Extend Plugin class
  2. Add resource loading methods
  3. Update PluginManager to use new features
  4. Document in plugin structure guide

Future Enhancements

  • HTTP/SSE transport for MCP servers
  • Plugin registry and discovery
  • Plugin versioning and updates
  • Plugin dependencies resolution
  • Caching and optimization
  • Multi-provider fallback
  • Plugin marketplace integration