Agent MCP Server

Introduction

MCP Agent is a powerful framework for building effective AI agents using the Model Context Protocol. It provides a set of composable patterns and tools that make it easier to create reliable, transparent, and effective AI agents.

Installation

You can install MCP Agent using pip:

pip install mcp-agent

Basic Usage

MCP Agent allows you to build agents using different patterns:

1. Using AugmentedLLM

AugmentedLLM is a core component that provides a unified interface to different LLM providers with enhanced capabilities:

from mcp_agent import OpenAIAugmentedLLM

# Initialize an AugmentedLLM
llm = OpenAIAugmentedLLM(model="gpt-4o")

# Generate text
response = llm.generate("Explain quantum computing in simple terms")

# Generate structured output
structured_response = llm.generate_structured(
    "List the top 3 programming languages",
    schema={
        "type": "array",
        "items": {
            "type": "object",
            "properties": {
                "name": {"type": "string"},
                "popularity": {"type": "string"}
            }
        }
    }
)

2. Building Agents with Workflows

MCP Agent provides workflow patterns for creating more complex agents:

from mcp_agent import Agent, Workflow, Step

# Define steps in your workflow
research_step = Step(
    name="research",
    description="Research information about a topic",
    agent=OpenAIAugmentedLLM(model="gpt-4o")
)

summarize_step = Step(
    name="summarize",
    description="Summarize the research findings",
    agent=OpenAIAugmentedLLM(model="gpt-4o")
)

# Create a workflow
research_workflow = Workflow(
    name="research_workflow",
    description="Research and summarize a topic",
    steps=[research_step, summarize_step]
)

# Execute the workflow
result = research_workflow.execute("Quantum computing")

3. Using the Orchestrator

For more complex agent behaviors, you can use the Orchestrator:

from mcp_agent import Orchestrator, Agent

# Create an orchestrator with multiple agents
orchestrator = Orchestrator(
    agents={
        "researcher": Agent(llm=OpenAIAugmentedLLM(model="gpt-4o")),
        "writer": Agent(llm=OpenAIAugmentedLLM(model="gpt-4o"))
    }
)

# Execute a task with the orchestrator
result = orchestrator.execute(
    "Research quantum computing and write a blog post about it",
    agent_selection_strategy="auto"
)

Model Providers

MCP Agent supports multiple model providers:

• OpenAI (GPT models)
• Anthropic (Claude models)
• Google (Gemini models)
• Custom model providers

You can select the appropriate AugmentedLLM implementation based on your preferred provider:

from mcp_agent import OpenAIAugmentedLLM, AnthropicAugmentedLLM, GoogleAugmentedLLM

openai_llm = OpenAIAugmentedLLM(model="gpt-4o")
anthropic_llm = AnthropicAugmentedLLM(model="claude-3-opus-20240229")
google_llm = GoogleAugmentedLLM(model="gemini-1.5-pro")

Advanced Features

Streaming Responses

MCP Agent supports streaming responses for real-time interaction:

for chunk in llm.generate_stream("Explain quantum computing step by step"):
    print(chunk, end="", flush=True)

Structured Output Generation

Generate JSON or other structured outputs with schema validation:

result = llm.generate_structured(
    "Extract entities from this text: John met Sarah at Starbucks on Tuesday.",
    schema={
        "type": "object",
        "properties": {
            "people": {"type": "array", "items": {"type": "string"}},
            "places": {"type": "array", "items": {"type": "string"}},
            "time": {"type": "string"}
        }
    }
)

HTTP Server Integration

MCP Agent can be used to create MCP-compatible HTTP servers:

from mcp_agent.server import MCPServer
from mcp_agent import OpenAIAugmentedLLM

# Create a server with an AugmentedLLM
server = MCPServer(llm=OpenAIAugmentedLLM(model="gpt-4o"))

# Start the server
server.start(host="0.0.0.0", port=8000)

Examples

For more detailed examples, check out the examples directory in the GitHub repository, which includes:

• Basic agent implementations
• Workflow patterns
• Model provider integrations
• HTTP server implementations
• Use cases for different domains

Additional Resources

• Model Context Protocol Documentation
• Building Effective Agents Research
• MCP Agent GitHub Repository