it_asset_management_mcp_adk_chat_agent by SudhakarC-hub - MCP Server

IT Asset Manager - ADK + MCP POC

A production-ready IT Asset Management system demonstrating Google ADK integration with Model Context Protocol (MCP) using Streamable HTTP (Pattern 2).

Architecture

User (Terminal)
    ↓
ADK Agent (Ollama Mistral)
    ↓
MCP Toolset (Streamable HTTP)
    ↓
MCP Server (Port 8002)
    ↓
FastAPI Backend (Port 8005)

Components

1. FastAPI Backend (`swagger_api/api_server.py`)

REST API for IT asset management
Endpoints: GET/POST /assets
Port: 8005

2. MCP Server (`mcp_servers/mcp_server.py`)

Exposes FastAPI as MCP tools using Streamable HTTP (production pattern)
Tools: list_assets, add_asset
Port: 8002
Endpoint: /mcp

3. ADK Agent (`it_asset_manager_app/agent.py`)

Google ADK agent with Ollama Mistral LLM
Uses McpToolset with StreamableHTTPConnectionParams
Connects to MCP server for tool execution

Setup

Prerequisites

Python 3.11+
Ollama with Mistral model
Virtual environment

Installation

# 1. Activate virtual environment
source venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Ensure Ollama is running
ollama run mistral

Running the System

Open 3 separate terminals:

Terminal 1: FastAPI Server

cd /Users/sudhakarchigurupati/ADKProject/mcp_project
source venv/bin/activate
python swagger_api/api_server.py

✅ Running on: http://localhost:8005

Terminal 2: MCP Server

cd /Users/sudhakarchigurupati/ADKProject/mcp_project
source venv/bin/activate
python mcp_servers/mcp_server.py

✅ Running on: http://localhost:8002/mcp

Terminal 3: ADK Agent

Option A: Terminal Chat (ADK CLI)

cd /Users/sudhakarchigurupati/ADKProject/mcp_project
source venv/bin/activate
adk run it_asset_manager_app

Option B: Web Chat (Streamlit)

cd /Users/sudhakarchigurupati/ADKProject/mcp_project
source venv/bin/activate
streamlit run streamlit_app/app.py

Opens in browser at http://localhost:8501

Usage

Once the ADK agent is running, try these commands:

You: hello
You: List all assets
You: Add asset id 10, name Surface Pro, owner Mike, status active
You: Show me the inventory

Key Files

mcp_project/
├── swagger_api/
│   └── api_server.py          # FastAPI backend
├── mcp_servers/
│   └── mcp_server.py          # MCP server (Streamable HTTP)
├── it_asset_manager_app/
│   ├── agent.py               # ADK agent with MCP toolset
│   ├── __init__.py
│   └── .env                   # Environment variables
├── requirements.txt           # Python dependencies
└── README.md                  # This file

How It Works

User Input → ADK agent receives user message
LLM Decision → Ollama Mistral decides which tool to use
MCP Call → ADK's McpToolset calls MCP server via Streamable HTTP
Tool Execution → MCP server executes tool (calls FastAPI)
Response → Result flows back through the chain to user

MCP Tools Available

`list_assets()`

Retrieves all IT assets from the inventory.

Example: "List all assets"

`add_asset(asset_id, name, owner, status)`

Adds a new IT asset to the inventory.

Parameters:

asset_id (int): Unique identifier
name (str): Asset name/model
owner (str): Person assigned to asset
status (str): Status (default: "active")

Example: "Add asset id 5, name iPad Pro, owner Sarah, status active"

Production Pattern

This implementation uses Pattern 2: Remote MCP Servers (Streamable HTTP) from the ADK MCP documentation, which is recommended for production deployments due to:

✅ Scalability (stateless mode)
✅ HTTP-based communication
✅ Suitable for Cloud Run / containerized deployments
✅ Better error handling and monitoring

Verification

Test FastAPI

curl http://localhost:8005/assets

Test MCP Server

The MCP server should show connection logs when the agent connects.

Test Agent

The agent should respond to greetings and execute MCP tools when asked.

Troubleshooting

Agent not responding

Check Ollama is running: ollama list
Verify MCP server is running on port 8002
Check FastAPI is running on port 8005

Port conflicts

# Check ports
lsof -i :8005  # FastAPI
lsof -i :8002  # MCP
lsof -i :11434 # Ollama

# Kill if needed
lsof -ti:8002 | xargs kill -9

Import errors

Make sure you're in the project root when running adk run.

Technology Stack

Google ADK: Agent framework
MCP (Model Context Protocol): Tool integration protocol
FastAPI: Backend API
Ollama: Local LLM (Mistral)
Streamable HTTP: Production-ready MCP transport
Python 3.11: Runtime
ADK Documentation
MCP Tools Guide
Model Context Protocol

Advanced: Using Multiple MCP Servers

ADK supports connecting to multiple MCP servers simultaneously, allowing you to build multi-service AI agents.

Use Case

Imagine you have separate services for different domains:

IT Asset Management (Port 8002)
HR Management (Port 8003)
Finance (Port 8004)

Each service has its own MCP server exposing domain-specific tools.

Implementation

from google.adk.agents.llm_agent import Agent
from google.adk.models.lite_llm import LiteLlm
from google.adk.tools.mcp_tool import McpToolset, StreamableHTTPConnectionParams


# MCP Server 1: IT Asset Management
asset_mcp = McpToolset(
    connection_params=StreamableHTTPConnectionParams(
        url="http://localhost:8002/mcp"
    )
)

# MCP Server 2: HR Management
hr_mcp = McpToolset(
    connection_params=StreamableHTTPConnectionParams(
        url="http://localhost:8003/mcp"
    )
)

# MCP Server 3: Finance
finance_mcp = McpToolset(
    connection_params=StreamableHTTPConnectionParams(
        url="http://localhost:8004/mcp"
    )
)

# Create agent with ALL MCP toolsets
root_agent = Agent(
    model=LiteLlm(
        model="ollama_chat/mistral:latest",
        api_base="http://localhost:11434"
    ),
    name="multi_service_agent",
    description="Multi-service agent managing IT, HR, and Finance",
    instruction="""You are a multi-service assistant with access to:
    
    IT Asset Management:
    - list_assets, add_asset
    
    HR Management:
    - list_employees, add_employee, get_employee_details
    
    Finance:
    - list_invoices, create_invoice, get_budget
    
    Use the appropriate tools based on user requests.""",
    tools=[
        asset_mcp,    # All IT asset tools
        hr_mcp,       # All HR tools
        finance_mcp   # All finance tools
    ],
)

Architecture

User: "List all assets and show me John's employee details"
    ↓
ADK Agent (Single LLM)
    ├─→ Asset MCP Server (Port 8002) → list_assets()
    └─→ HR MCP Server (Port 8003) → get_employee_details("John")

Benefits

✅ Separation of Concerns - Each service has its own MCP server
✅ Independent Scaling - Scale each service separately
✅ Team Ownership - Different teams can own different MCP servers
✅ Easy Integration - Just add another McpToolset to connect a new service
✅ Single Agent Interface - Users interact with one agent that orchestrates all services

How It Works

Tool Discovery: ADK automatically discovers all tools from all MCP servers
LLM Decision: The LLM analyzes the user request and decides which tools to use
Parallel Execution: ADK can call multiple tools from different MCP servers in parallel
Response Synthesis: The LLM combines results from multiple services into a coherent response

Production Considerations

Service Discovery: Use environment variables for MCP server URLs
Health Checks: Implement health check endpoints for each MCP server
Error Handling: Each MCP server should handle errors independently
Authentication: Add authentication to MCP servers in production
Rate Limiting: Implement rate limiting per MCP server
Monitoring: Monitor each MCP server separately

Example: Environment-Based Configuration

import os

# Load MCP server URLs from environment
ASSET_MCP_URL = os.getenv("ASSET_MCP_URL", "http://localhost:8002/mcp")
HR_MCP_URL = os.getenv("HR_MCP_URL", "http://localhost:8003/mcp")
FINANCE_MCP_URL = os.getenv("FINANCE_MCP_URL", "http://localhost:8004/mcp")

# Create toolsets
asset_mcp = McpToolset(
    connection_params=StreamableHTTPConnectionParams(url=ASSET_MCP_URL)
)
hr_mcp = McpToolset(
    connection_params=StreamableHTTPConnectionParams(url=HR_MCP_URL)
)
finance_mcp = McpToolset(
    connection_params=StreamableHTTPConnectionParams(url=FINANCE_MCP_URL)
)

# Agent with all services
root_agent = Agent(
    model=LiteLlm(model="ollama_chat/mistral:latest"),
    tools=[asset_mcp, hr_mcp, finance_mcp],
)

This pattern allows you to build enterprise-grade multi-service AI agents that can orchestrate across your entire organization's services! 🚀

SudhakarC-hub/it_asset_management_mcp_adk_chat_agent

IT Asset Manager - ADK + MCP POC

Architecture

Components

1. FastAPI Backend (swagger_api/api_server.py)

2. MCP Server (mcp_servers/mcp_server.py)

3. ADK Agent (it_asset_manager_app/agent.py)

Setup

Prerequisites

Installation

Running the System

Terminal 1: FastAPI Server

Terminal 2: MCP Server

Terminal 3: ADK Agent

Usage

Key Files

How It Works

MCP Tools Available

list_assets()

add_asset(asset_id, name, owner, status)

Production Pattern

Verification

Test FastAPI

Test MCP Server

Test Agent

Troubleshooting

Agent not responding

Port conflicts

Import errors

Technology Stack

Advanced: Using Multiple MCP Servers

Use Case

Implementation

Architecture

Benefits

How It Works

Production Considerations

Example: Environment-Based Configuration

1. FastAPI Backend (`swagger_api/api_server.py`)

2. MCP Server (`mcp_servers/mcp_server.py`)

3. ADK Agent (`it_asset_manager_app/agent.py`)

`list_assets()`

`add_asset(asset_id, name, owner, status)`