Skip to main content
Glama
MCP Server with Tool, Resource, Prompt, and Sampling

MCP Server with Tool, Resource, Prompt, and Sampling

by Pkaran26

MCP Server with Tool, Resource, Prompt, and Sampling

A full-featured Model Context Protocol (MCP) server built with Node.js and TypeScript. This project demonstrates how to implement and integrate the core capabilities of MCP into a single server, communicating over stdio.

πŸš€ Features

This server showcases the four main pillars of the Model Context Protocol:

  • Resources (users://all): Exposes local data to the LLM. It safely reads from src/data/users.json using node:fs so the LLM can analyze existing user data.

  • Tools (create-user): Allows the LLM to take action by securely writing new user data to the local file system with strict validation via zod.

  • Sampling (create-random-user): Demonstrates server-to-client LLM requests. When triggered, the server uses the sampling/createMessage endpoint to ask the client LLM to generate fake user data, parses the response, and automatically saves it.

  • Prompts: Configured to support structured prompt templates that guide the LLM's interactions with the provided tools and resources.

Related MCP server: LocalData MCP Server

πŸ› οΈ Prerequisites

  • Node.js: v20 or higher (Tested on v24)

  • Package Manager: yarn or npm

πŸ“¦ Installation

  1. Clone the repository and navigate to the directory:

    cd mcp-server-sampling

    Install dependencies:

    yarn install

    (Ensure @modelcontextprotocol/sdk and zod are installed).

Initialize the data file: Ensure you have an empty or valid JSON array in src/data/users.json:

JSON:πŸ—οΈ Build & Run

Because this project uses modern ES Modules (ESM) and is strictly typed, you must build the project before connecting it to an MCP client like VS Code.

Compile the TypeScript code to JavaScript:

Bash yarn build (This outputs the compiled code to the /build directory).

(Optional) Test the server locally using the MCP Inspector:

npx @modelcontextprotocol/inspector node build/server.js

πŸ”Œ VS Code Configuration

To connect this server to VS Code (or Cursor/Claude Desktop), you need to register it in your client's MCP settings.

For VS Code, update your configuration (usually found in your MCP settings file) to point to the compiled server.js file:

{
  "mcp.servers": {
    "my-mcp-server-sampling": {
      "type": "stdio",
      "command": "node",
      "args": ["${workspaceFolder}/build/server.js"],
      "cwd": "${workspaceFolder}"
    }
  }
}

⚠️ Important Troubleshooting Note:

Always remember to run yarn build after making any changes to src/server.ts. VS Code executes the compiled build/server.js file, so if you skip the build step, your changes will not be reflected in the IDE.

πŸ› Common Pitfalls Avoided

ESM Imports: All @modelcontextprotocol/sdk imports include the .js extension to comply with strict Node.js ESM resolution.

Stdio Corruption: Uses console.error exclusively for logging to prevent corrupting the JSON-RPC stdout stream used for client-server communication.

Dynamic JSON Loading: Uses node:fs with absolute paths rather than import statements to reliably read users.json both in development (tsx) and production (node).

Available Tools

2 tools
create-random-userA

Create a random user with fake data

Inputs: no named parameters.

create-userB

Create a new user

Inputs: name (required), email (required), phone (required), address.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    B3.4/5.0

    Scored across 2 tools

    Disambiguation4/5

    Both tools create users, but one creates a specific user and the other creates a random one. Descriptions clearly distinguish them, though the core action is the same.

    Naming Consistency5/5

    Both tools follow a consistent verb_noun snake_case pattern: create-user and create-random-user.

    Tool Count2/5

    With only 2 tools focused solely on user creation, the count is too low for a server that also advertises resources, prompts, and sampling capabilities.

    Completeness2/5

    The tools only cover creation (specific and random) but lack essential user management operations like reading, updating, deleting, or listing users.

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      Gives LLM agents access to local and remote data via databases, files, graphs, and structured documents, along with a full data science toolkit for analysis and modeling.
      15
      Apache 2.0
    • A
      license
      Not graded
      quality
      D
      maintenance
      Enables AI agents with long-term memory and retrieval-augmented generation (RAG) capabilities, allowing them to recall past conversations, search local files, and learn user preferences.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI assistants to persist and retrieve user information across chats via a local knowledge graph with entities, relations, and observations, using a SQLite-backed MCP interface.
      3 npm
      Apache 2.0