Bedrock Inline AgentでPerplexityなど複数のMCPサーバを使うサンプルを試してみる

git clone https://github.com/awslabs/amazon-bedrock-agent-samples.git
cd amazon-bedrock-agent-samples/src/InlineAgent

python -m venv .venv
source .venv/bin/activate

python -m pip install -e .

InlineAgent_hello us.anthropic.claude-3-5-haiku-20241022-v1:0

[bedrock]
aws_access_key_id = {Your Access Key}
aws_secret_access_key = {Your Secret Access Key}
region = us-east-1

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "bedrock:*"
            ],
            "Resource": "*"
        }
    ]
}

await InlineAgent(
    foundation_model="us.anthropic.claude-3-5-sonnet-20241022-v2:0",
    instruction="""You are a friendly assistant that is responsible for resolving user queries.
    
    You have access to search, cost tool and code interpreter. 
    
    """,
    agent_name="cost_agent",
    action_groups=[
        cost_action_group,
        {
            "name": "CodeInterpreter",
            "builtin_tools": {
                "parentActionGroupSignature": "AMAZON.CodeInterpreter"
            },
        },
    ],
    profile="bedrock",  # 追加

Input Tokens: 600 Output Tokens: 137
Thought: The user has greeted me and asked about my capabilities. I'll respond in a friendly manner and use the user interaction tool to engage with them.
Hello there! I'm doing great, thank you for asking. I'm a friendly assistant who loves to say hello to everything! What would you like help with today? I'm ready to assist you with any questions or tasks you might have.
Agent made a total of 1 LLM calls, using 737 tokens (in: 600, out: 137), and took 10.5 total seconds

env | grep AWS

cd examples/mcp/cost_explorer_agent

AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_REGION=us-east-1
BEDROCK_LOG_GROUP_NAME=
PERPLEXITY_API_KEY=

git clone https://github.com/aws-samples/sample-cloud-spend-mcp-server.git
cd sample-cloud-spend-mcp-server
docker build -t aws-cost-explorer-mcp .

await InlineAgent(
    foundation_model="us.anthropic.claude-3-5-sonnet-20241022-v2:0",
    instruction="""You are a friendly assistant that is responsible for resolving user queries.
    
    You have access to search, cost tool and code interpreter. 
    
    """,
    agent_name="cost_agent",
    action_groups=[
        cost_action_group,
        {
            "name": "CodeInterpreter",
            "builtin_tools": {
                "parentActionGroupSignature": "AMAZON.CodeInterpreter"
            },
        },
    ],
    profile="bedrock", # 追加

python main.py

% python main.py
Processing request of type ListToolsRequest

Connected to server with tools:['get_bedrock_daily_usage_stats', 'get_bedrock_hourly_usage_stats', 'get_ec2_spend_last_day', 'get_detailed_breakdown_by_day']
Processing request of type ListToolsRequest
Processing request of type ListToolsRequest
Perplexity Ask MCP Server running on stdio

Connected to server with tools:['perplexity_ask']
SessionId: 94a1b11f-c1f7-4a23-9a72-36a60fb40d4f
2025-04-11 23:50:35,062 - botocore.credentials - INFO - Found credentials in shared credentials file: ~/.aws/credentials

Input Tokens: 1502 Output Tokens: 155
Thought: Let me help you understand Amazon Bedrock and analyze your AWS spending. I'll:
1. Use Perplexity to get accurate information about Amazon Bedrock
2. Get detailed cost breakdown for the last 7 days
3. Create a visualization of the spending data

...

Code interpreter output: 
Code Interpreter created new files



Files saved in output directory                                                                                    
Input Tokens: 16037 Output Tokens: 217
Thought: Now I can provide a comprehensive answer about Amazon Bedrock and the cost analysis.
Amazon Bedrock is AWS's fully managed service that provides access to foundation models (FMs) from leading AI companies through a unified API. It allows developers to build and scale generative AI applications without managing infrastructure.

Looking at your AWS spending over the last 7 days, the top services by cost were:
1. Claude 3.5 Sonnet (Amazon Bedrock Edition): $2.26
2. AWS Security Hub: $1.62
3. AmazonCloudWatch: $0.24
4. Amazon GuardDuty: $0.16
5. AWS Config: $0.14

I've created a bar chart showing the cost distribution across these services. The visualization clearly shows that Amazon Bedrock (Claude 3.5 Sonnet) and AWS Security Hub were your highest spending services during this period.


Files saved in output directory                                                                                    

Agent made a total of 5 LLM calls, using 38789 tokens (in: 37886, out: 903), and took 54.7 total seconds