AISuite MCP: Multi-Component Platform for AI-to-AI Peer Review

AISuite MCP extends the AISuite library to enable AI-to-AI peer review through structured prompt-response frameworks. This platform facilitates advanced reasoning techniques such as chain-of-thought, self-consistency, and debate-style AI interactions, while allowing the originating AI to specify its desired response format.

Features

• AI-to-AI Peer Review: Have one AI generate a response and another AI critique and improve it
• Flexible Review Mechanisms: Support for standard review, debate mode, and self-consistency checks
• Format Specification: Originating AI can specify desired response formats and structures
• Tool Integration: Transparent access to tools for verification and enhanced reasoning
• Asynchronous Processing: Support for both synchronous and asynchronous workflows

Installation

From PyPI

pip install aisuite-mcp

From Source

# Clone the repository
git clone https://github.com/microsoftmj/aisuitemcp.git
cd aisuitemcp

# Install in development mode
pip install -e .

Prerequisites

This package requires:

• Python 3.9 or later
• AISuite 0.1.5 or later
• Valid API keys for LLM providers (OpenAI, Anthropic, etc.)

Quick Start

from aisuite_mcp import MCPClient, ReviewSpec, ReviewType

# Initialize the client
mcp_client = MCPClient(
    provider_configs={
        "openai": {"api_key": "your_openai_key"},
        "anthropic": {"api_key": "your_anthropic_key"}
    }
)

# Define review specification
review_spec = ReviewSpec(
    generator_model="openai:gpt-4",
    reviewer_model="openai:gpt-4",
    review_type=ReviewType.STANDARD,
    review_criteria=["factual_accuracy", "reasoning", "completeness"],
    response_format={
        "structure": "markdown",
        "sections": ["reasoning", "answer", "confidence"]
    }
)

# Get peer-reviewed response
response = mcp_client.create_reviewed_completion(
    prompt="Explain the implications of quantum computing on cryptography",
    review_spec=review_spec
)

print(response.final_answer)

Architecture

AISuite MCP is built as a wrapper around AISuite, leveraging its unified interface to multiple LLM providers while adding multi-component orchestration capabilities. The architecture consists of:

• MCP Orchestrator: Coordinates the workflow between different AI models and components
• Format Parser: Handles parsing and enforcement of response format specifications
• Review Manager: Implements various review strategies and processes
• Tools Manager: Manages tool registration, execution, and result integration

Format Specification

The format specification allows the originating AI to define how it wants responses structured:

from aisuite_mcp import FormatSpec, ResponseFormat

format_spec = FormatSpec(
    structure=ResponseFormat.JSON,
    sections=["reasoning", "answer", "confidence"],
    example='{ "reasoning": "...", "answer": "...", "confidence": 0.95 }'
)

Review Mechanisms

AISuite MCP supports multiple review mechanisms:

• Standard Review: One AI generates, another reviews based on criteria
• Debate Mode: Two AIs debate a question with multiple turns
• Self-Consistency: Generate multiple reasoning paths and determine consensus

Tool Integration

You can register custom tools that will be available during the review process:

def search(query: str) -> str:
    """Search for information on a topic."""
    # Implement search functionality
    return f"Results for: {query}"

# Register the tool
mcp_client.register_tool(search)

# Use in review specification
review_spec = ReviewSpec(
    generator_model="openai:gpt-4",
    reviewer_model="openai:gpt-4",
    tools=["search"]  # Make search tool available
)

Examples

See the examples/ directory for more detailed usage examples:

• basic_usage.py: Demonstrates core functionality
• test_implementation.py: Simple test script for verification

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT