How to Use MCPs and Tools Together in a Single Agent?

[daniel7byte]

I haven’t been able to get the MCPs and the tools to work together. In my tests, it either uses the MCPs or the tools, but never both for the tasks I need to perform. Should I set up an Agent2Agent configuration, with one agent using tools and another one connected to the MCPs?

from mcp import stdio_client, StdioServerParameters
from strands import Agent
from strands_tools import calculator, file_write, python_repl
from strands.tools.mcp import MCPClient

import os
os.environ["BYPASS_TOOL_CONSENT"] = "true"

# Connect to an MCP server using stdio transport
# Note: uvx command syntax differs by platform

# For macOS/Linux:
stdio_mcp_client_aws_documentation = MCPClient(lambda: stdio_client(
    StdioServerParameters(
        command="uvx",
        args=["awslabs.aws-documentation-mcp-server@latest"]
    )
))

stdio_mcp_client_aws_diagram = MCPClient(lambda: stdio_client(
    StdioServerParameters(
        command="uvx",
        args=["awslabs.aws-diagram-mcp-server"],
        env={
            "FASTMCP_LOG_LEVEL": "ERROR"
        }
    )
))

stdio_mcp_client_aws_pricing = MCPClient(lambda: stdio_client(
    StdioServerParameters(
        command="uvx",
        args=["awslabs.aws-pricing-mcp-server@latest"],
        env={
            "FASTMCP_LOG_LEVEL": "ERROR",
            "AWS_PROFILE": "default",
            "AWS_REGION": "us-east-1"
        }
    )
))

# Create an agent with MCP tools
with stdio_mcp_client_aws_documentation, stdio_mcp_client_aws_diagram, stdio_mcp_client_aws_pricing:
    # Get the tools from the MCP server
    tools = stdio_mcp_client_aws_documentation.list_tools_sync(), stdio_mcp_client_aws_diagram.list_tools_sync() + stdio_mcp_client_aws_pricing.list_tools_sync()


    # Create an agent with these tools
    agent = Agent(tools=[tools, calculator, file_write, python_repl]),
    result = agent("""
          What is AWS Lambda?

Create an AWS architecture diagram for a simple serverless API using Lambda, API Gateway, and another service (you can suggest one) to enable real-time communication. Also, connect the Lambda to a database like DynamoDB, with access to Amazon Bedrock for chatbot functionality using AI.

Generate the image and save the diagram as a file named diagrama_architectura.png. Use the official AWS diagramming tool. If generating the diagram is not possible, then give me the architecture in Python using the diagrams library.

Now, based on this architecture, provide an estimated cost using the AWS Pricing Calculator. Assume that around 5 chats are started daily, each with approximately 15 interactions (with 50% of the paragraphs coming from the user and 50% from the AI agent using Claude 3.5 Sonnet).
          """)
          

[zastrowm]

I believe this is due to the fact that the tools list here is nested lists of tools. I believe the syntax that works is:

mcp_tools = [*stdio_mcp_client_aws_documentation.list_tools_sync(), *stdio_mcp_client_aws_diagram.list_tools_sync(), *stdio_mcp_client_aws_pricing.list_tools_sync()]
tools=[*mcp_tools, calculator, file_write, python_repl]

agent = Agent(tools=tools),

If that is the case, then I think #50 would have been useful here