mcp.plugins

MCP Plugins Module.

This module provides plugin implementations for the unified MCP platform architecture. All plugins inherit from PluginPlatform and implement the Pydantic-first design pattern with intelligent inheritance from our Phase 1 base platform models.

Plugin Architecture

Base Inheritance Chain

  • BasePlatform (haive-dataflow) - Foundation with core capabilities

  • PluginPlatform (haive-dataflow) - Plugin-specific enhancements

  • MCPBrowserPlugin (haive-mcp) - Browser automation plugin

Key Features

  • Pydantic-First Design: Full type safety and validation

  • Intelligent Inheritance: Automatic capability composition

  • Phase 2 Implementation: Advanced plugin functionality

Available Plugins

Browser Plugin

  • MCPBrowserPlugin - Browser automation and web interaction

  • Features: Page navigation, element interaction, screenshot capture

  • Integration: Works with Playwright and Selenium

Plugin Registry

  • get_plugin_registry() - Get available plugin registry

  • get_plugin_class() - Load plugin class by name

Note

This is a Phase 2 implementation focusing on advanced plugin capabilities with full Pydantic integration and intelligent inheritance patterns.

Submodules

Classes

MCPBrowserPlugin

Plugin for managing our 63 downloaded MCP servers.

Package Contents

class mcp.plugins.MCPBrowserPlugin

Bases: haive.dataflow.platform.models.PluginPlatform

Plugin for managing our 63 downloaded MCP servers.

This plugin inherits from PluginPlatform and specializes in managing the MCP servers we successfully downloaded during our bulk download session. It demonstrates the intelligent inheritance pattern by extending platform capabilities while maintaining the Pydantic-first design philosophy.

Inheritance Features: - Inherits: platform_id, status, metadata, lifecycle methods from BasePlatform (via PluginPlatform) - Inherits: entry_point, routes, priorities, dependencies from PluginPlatform - Extends: MCP-specific server management and discovery capabilities - Specializes: Downloaded server data integration and caching

Real Integration: - Works with our actual mcp_servers_data.csv file - Uses our install report JSON files - Loads our 63 successfully downloaded servers - Provides web interface for browsing servers

Examples

Basic plugin creation:

plugin = MCPBrowserPlugin()
# Uses intelligent defaults for all inherited fields

With custom data paths:

plugin = MCPBrowserPlugin(
    servers_data_file=Path("custom/path/servers.csv"),
    install_reports_pattern="custom_install_*.json"
)

Plugin initialization:

await plugin.initialize()
servers = plugin.get_servers()
print(f"Loaded {len(servers)} downloaded servers")
async cleanup() → None

Cleanup plugin resources.

get_plugin_stats() → Dict[str, Any]

Get comprehensive plugin statistics.

Returns:

Dictionary with plugin statistics and server information

get_server_by_name(server_name: str) → haive.dataflow.platform.models.DownloadedServerInfo | None

Get a specific server by name.

Parameters:

server_name – Server name to search for

Returns:

Server if found, None otherwise

get_servers() → List[haive.dataflow.platform.models.DownloadedServerInfo]

Get our 63 downloaded servers with intelligent caching.

This method implements intelligent caching to avoid repeatedly loading and processing our server data. It uses the cache TTL to determine when to refresh the data.

Returns:

List of DownloadedServerInfo instances for our downloaded servers

Examples

>>> plugin = MCPBrowserPlugin()
>>> servers = plugin.get_servers()
>>> len(servers)
63
>>> all(s.source == ServerSource.DOWNLOADED for s in servers)
True
get_servers_by_language(language: str) → List[haive.dataflow.platform.models.DownloadedServerInfo]

Get servers filtered by programming language.

Parameters:

language – Programming language to filter by

Returns:

List of servers matching the language

get_servers_by_stars(min_stars: int = 0) → List[haive.dataflow.platform.models.DownloadedServerInfo]

Get servers with at least the specified number of stars.

Parameters:

min_stars – Minimum number of GitHub stars

Returns:

List of servers sorted by star count (descending)

async initialize() → None

Initialize plugin with validation and server loading.

This method extends the base PluginPlatform initialization with MCP-specific setup and validation.

register_routes(app: fastapi.FastAPI) → None

Register plugin routes with FastAPI app.

This method creates FastAPI routes for browsing our downloaded servers. It’s called by the platform during plugin initialization.

Parameters:

app – FastAPI application instance to register routes with

classmethod validate_servers_data_file_exists(v: pathlib.Path) → pathlib.Path

Validate that servers data file exists (when not using defaults).

cache_timestamp: datetime.datetime | None = None
cache_ttl_seconds: int = None
cached_servers: List[haive.dataflow.platform.models.DownloadedServerInfo] | None = None
description: str = None
downloaded_servers_path: pathlib.Path = None
entry_point: str = None
install_reports_pattern: str = None
model_config
platform_id: str = None
platform_name: str = None
priority: int = None
provides_discovery: bool = None
provides_health_checks: bool = None
provides_servers: bool = None
router: fastapi.APIRouter | None = None
routes_prefix: str = None
servers_data_file: pathlib.Path = None
supports_discovery: bool = None
supports_health_monitoring: bool = None