	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Agentic AI > MCP Protocol > MCP Server with Resources and Prompts

MCP Server with Resources and Prompts

Author: Venkata Sudhakar

MCP defines three primitives: Tools (executable functions), Resources (readable data), and Prompts (reusable prompt templates). Most tutorials focus on tools, but Resources and Prompts unlock powerful patterns. Resources let the MCP server expose live data - like a product catalogue or policy document - that the agent reads as context. Prompts let the server provide pre-built instruction templates that ensure consistent agent behaviour across different sessions.

FastMCP exposes resources with @mcp.resource() and prompts with @mcp.prompt(). Resources have a URI scheme (e.g. shopmax://policies/returns) and return text content. Prompts accept parameters and return formatted message strings. Both are loaded by the ADK agent client and can be injected into the agent instruction or used as context before a query.

The below example shows a ShopMax India MCP server that exposes a returns policy resource and a complaint response prompt template.

# shopmax_resources_server.py
from mcp.server.fastmcp import FastMCP
mcp = FastMCP("shopmax-resources")

# Resource: static policy document
@mcp.resource("shopmax://policies/returns")
def get_returns_policy() -> str:
    # Returns the ShopMax India return policy as text
    return (
        "ShopMax India Return Policy\n"
        "================================\n"
        "1. Electronics: 7 days from delivery if defective.\n"
        "2. Clothing: 30 days, unworn with tags attached.\n"
        "3. Appliances: 14 days if defective, manufacturer warranty applies.\n"
        "4. Non-returnable: Software, consumables, personalised items.\n"
        "Refund credited within 3-5 business days to original payment method.\n"
    )

# Resource: live product count
@mcp.resource("shopmax://inventory/summary")
def get_inventory_summary() -> str:
    # Returns current inventory summary
    return (
        "ShopMax Inventory - April 2026\n"
        "Laptops: 42 units | TVs: 15 units | Phones: 120 units\n"
        "Total SKUs: 3 | Total Units: 177\n"
    )

# Prompt: complaint response template
@mcp.prompt()
def complaint_response_prompt(customer_name: str, issue: str, resolution: str) -> str:
    # Returns a formatted complaint response template
    return (
        f"Dear {customer_name},\n\n"
        f"Thank you for reaching out to ShopMax India regarding: {issue}.\n"
        f"We sincerely apologise for the inconvenience caused.\n\n"
        f"Resolution: {resolution}\n\n"
        "We value your loyalty and hope to serve you better.\n"
        "Warm regards,\nShopMax India Customer Care"
    )

# Tool: fetch resource and prompt
@mcp.tool()
def get_policy_text(policy_name: str) -> str:
    # Retrieve a policy document by name
    policies = {"returns": get_returns_policy(), "inventory": get_inventory_summary()}
    return policies.get(policy_name, "Policy not found")

@mcp.tool()
def generate_complaint_response(customer_name: str, issue: str, resolution: str) -> str:
    # Generate a formatted complaint response
    return complaint_response_prompt(customer_name, issue, resolution)

if __name__ == "__main__":
    mcp.run(transport="stdio")

Connect an ADK agent to use the resources and prompt templates,

import asyncio
from google.adk.agents import LlmAgent
from google.adk.tools.mcp_tool.mcp_toolset import MCPToolset, StdioServerParameters
from google.adk.sessions import InMemorySessionService
from google.adk.runners import Runner
from google.genai import types

async def main():
    mcp_tools, exit_stack = await MCPToolset.from_server(
        connection_params=StdioServerParameters(command="python", args=["shopmax_resources_server.py"])
    )

agent = LlmAgent(
        name="support_agent",
        model="gemini-2.0-flash",
        tools=mcp_tools,
        instruction="You are ShopMax India support. Use tools to retrieve policies and generate responses."
    )

session_service = InMemorySessionService()
    runner = Runner(agent=agent, app_name="support", session_service=session_service)
    session = session_service.create_session(app_name="support", user_id="agent1")

queries = [
        "What is the return policy for electronics?",
        "Generate a complaint response for Meera Singh about a delayed TV delivery. Resolution: full refund of Rs 35000.",
    ]
    for q in queries:
        msg = types.Content(role="user", parts=[types.Part(text=q)])
        async for event in runner.run_async(user_id="agent1", session_id=session.id, new_message=msg):
            if event.is_final_response():
                print("Q:", q[:60] + "...")
                print("A:", event.content.parts[0].text)
                print()

await exit_stack.aclose()

asyncio.run(main())

It gives the following output,

Q: What is the return policy for electronics?...
A: Electronics at ShopMax India are returnable within 7 days from delivery
if defective. Refunds are credited within 3-5 business days to the original
payment method.

Q: Generate a complaint response for Meera Singh about a delayed...
A: Dear Meera Singh,

Thank you for reaching out to ShopMax India regarding: delayed TV delivery.
We sincerely apologise for the inconvenience caused.

Resolution: full refund of Rs 35000

We value your loyalty and hope to serve you better.
Warm regards,
ShopMax India Customer Care

Resources are ideal for content that changes infrequently - policy documents, pricing tiers, or configuration. Update the resource implementation when the underlying data changes and all agents get the new content automatically without any agent code changes. Prompts standardise communication tone across your entire agent fleet: whether 10 or 100 agents are running, they all use the same complaint_response_prompt template, ensuring brand-consistent customer communications. Version your prompt templates using the prompt name (e.g. complaint_response_v2) to allow gradual rollout of tone changes.

Send your comments, suggestions or queries regarding this site to [email protected].