Understanding MCP Resources

•3 min read
mcpairesourcesengineering

Understanding MCP Resources

Resources represent data or files that an MCP client can read. This includes text, JSON, binary data, images, or any file content. This post explores how resources work using the SQLite MCP server as a case study.

TL;DR

Resources are data or files that can be read. Tools are functions that can be executed. The SQLite MCP server exposes table schemas as resources and provides a query tool. Models read the schemas, then write queries using the tool.

Case Study: SQLite MCP Server

The SQLite MCP server provides a clear example of resources in practice. You point it at any SQLite database file, and it exposes the schemas as resources.

What it exposes:

Resources:

  • schema://sqlite/{table} - The CREATE TABLE statement for any table in the database
  • schema://sqlite/all - JSON mapping of all table schemas

Tool:

  • sql_query - Executes read-only SQL SELECT statements

How it works:

python
# Using the official MCP Python SDK
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
 
server_params = StdioServerParameters(
    command="uvx",
    args=["mcp-server-sqlite",
          "--db-path", "my_database.db"]
)
 
async with stdio_client(server_params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
 
        # 1. List available resources
        resources = await session.list_resources()
        # Returns URIs for whatever tables exist:
        # schema://sqlite/products
        # schema://sqlite/customers
        # schema://sqlite/all
 
        # 2. Read a schema resource
        result = await session.read_resource(
            uri="schema://sqlite/products"
        )
        schema = result.contents[0].text
        # Returns the actual CREATE TABLE statement:
        # CREATE TABLE products (
        #   id INTEGER PRIMARY KEY,
        #   name TEXT,
        #   price REAL
        # )
 
        # 3. Model sees the schema structure
 
        # 4. Model calls the query tool
        data = await session.call_tool(
            "sql_query",
            {
                "query": "SELECT * FROM products WHERE price < 100"
            }
        )

The model reads schema resources to understand table structure, then writes precise queries.

Resources vs Tools

Resources are data that can be read:

  • Database table schemas
  • Documentation files
  • Configuration data
  • Dataset descriptions
  • Binary files, images

Tools are functions that can be executed:

  • SQL queries
  • File searches
  • API calls
  • Calculations

Resources provide information. Tools perform actions.

Current State

As of November 2025, Claude Desktop requires manual resource selection. Users click the plus icon to add resources to conversations. The MCP specification describes "automatic context inclusion based on the AI model's selection," but this is not widely implemented yet.

When to Use Resources

Use resources when you have data or files that models should read:

  • Database schemas - Table structures that guide query construction
  • Documentation - API references, guides, explanations
  • Configuration - Settings, constants, available options
  • Dataset descriptions - What data exists, its structure, date ranges
  • Files - Text, JSON, binary data, images, any file content

Use tools when you have operations to execute:

  • Queries - Search, filter, aggregate data
  • Actions - Create, update, delete operations
  • Computations - Calculate, transform, analyze
  • External calls - Fetch URLs, call APIs

References