complete_mcp_with_parent_retriever¶

Complete End-to-End MCP Example with Parent Document Retriever.

This example demonstrates: 1. Parent-child document retrieval for MCP servers 2. Self-query capabilities with metadata filtering 3. HITL approval for server installation 4. Live server testing with tools, resources, and prompts 5. Hierarchical categorization with AutoTree

The parent-child pattern allows: - Small chunks for precise similarity search (capabilities, keywords) - Full server documentation returned for context - Metadata filtering (stars, category, etc.) via self-query

Classes¶

`HITLRequest`	HITL approval request.
`MCPServerInfo`	Complete MCP server information.
`MCPSystemWithParentRetriever`	MCP system using parent-child document retrieval.

Functions¶

main()

Run the complete end-to-end example.

Module Contents¶

class complete_mcp_with_parent_retriever.HITLRequest(/, **data: Any)¶

Bases: pydantic.BaseModel

HITL approval request.

action: str¶

details: dict[str, Any] = None¶

risk_level: str = 'medium'¶

server_name: str¶

class complete_mcp_with_parent_retriever.MCPServerInfo(/, **data: Any)¶

Bases: pydantic.BaseModel

Complete MCP server information.

capabilities: list[str] = None¶

category: str¶

description: str¶

install_command: str | None = None¶

name: str¶

repository_url: str¶

setup_instructions: str | None = None¶

stars: int = 0¶

transport_types: list[str] = None¶

class complete_mcp_with_parent_retriever.MCPSystemWithParentRetriever(engine: haive.core.engine.aug_llm.AugLLMConfig)¶

MCP system using parent-child document retrieval.

async demonstrate_retrieval() → None¶: Demonstrate different retrieval patterns.

async install_top_server_with_hitl() → str | None¶: Find and install top server with HITL approval.

async setup_retrievers() → None¶: Set up both parent-child and self-query retrievers.

async test_installed_server(server_name: str) → None¶: Test installed server with all MCP capabilities.

doc_loader¶

engine¶

installed_servers: dict[str, dict]¶

parent_retriever = None¶

rag_agent = None¶

self_query_retriever = None¶

async complete_mcp_with_parent_retriever.main()¶: Run the complete end-to-end example.