Go back

MCP Servers: how to extend LLM capabilities with real tools

November 30, 2025
Anonymous
2 Views
Featured
AI & Machine Learning
Disclaimer: This text was translated from Polish using AI and has not been human-verified. Differences between the Polish and English versions may exist.
Table of contents
1. Revolution in working with AI: how to extend the capabilities of language models
2. What are MCP servers?
2.1. The N×M problem that MCP solves
2.2. How does it work in practice?
3. Why use MCP servers?
3.1. Elimination of hallucinations and access to up-to-date data
3.2. Automation of complex workflows
3.3. Standardization and reduction of development costs
3.4. Security and access control
4. How to add MCP servers to LLMs?
4.1. Claude Desktop — simplest configuration
4.2. ChatGPT — Developer Mode with full MCP support
4.3. Other tools supporting MCP
5. Best MCP servers for developers
5.1. GitHub MCP — code management without leaving the editor
5.2. Filesystem MCP — intelligent file management
5.3. Docker MCP — containerization from chat
5.4. Sequential Thinking MCP — AI that thinks out loud
5.5. PostgreSQL MCP — databases in natural language
5.6. Context7 MCP — up-to-date documentation always at hand
5.7. AWS and Azure MCP — cloud under control
5.8. Playwright MCP — UI test automation
5.9. Microsoft Learn Docs MCP — always up-to-date .NET documentation
5.10. Git MCP — version control from AI
6. Best MCP servers for non-technical users
6.1. Google Drive MCP — intelligent document management
6.2. Slack MCP — team communication without switching windows
6.3. Notion MCP — knowledge and task management
6.4. Gmail and Google Calendar MCP — productivity to the max
6.5. Brave Search MCP — private web search
6.6. Figma MCP — design meets development
6.7. Excel and Google Sheets MCP — data analytics simplified
6.8. Puppeteer MCP — browser automation
6.9. YouTube MCP — channel management and analytics
6.10. Google Workspace MCP — everything in one place
7. Docker MCP Toolkit — managing MCP servers in one place
8. Security and best practices
8.1. Current MCP security challenges
8.2. How to protect yourself against threats?
9. The Future of Model Context Protocol
9.1. Asynchronous operations
9.2. MCP Registry in production
9.3. Automatic server detection
9.4. Horizontal scaling
9.5. Support in Gemini
10. Key resources and documentation
11. Summary

1. Revolution in working with AI: how to extend the capabilities of language models



Language models, such as Claude or ChatGPT, have revolutionized the way we work with computers. However, despite impressive capabilities, these tools have a significant limitation — they are isolated from real data and systems that we use every day. They cannot search your drive themselves, update a database, or interact with the APIs of external services. At least until now.

In November 2024, Anthropic presented the Model Context Protocol (MCP) — an open standard that changes the rules of the game. MCP acts as a universal interface that allows AI models to connect with external data sources and tools in a standardized and safe way. It is like giving your AI assistant real hands with which it can perform specific tasks.

Already in March 2025, OpenAI officially adopted MCP into its ecosystem, and Google DeepMind announced support for MCP in upcoming versions of Gemini. Today, MCP is becoming an industry standard — its adoption by the biggest players in the AI market confirms that this is not a temporary trend, but the future of interaction with intelligent assistants.

In this article, I will show what MCP servers are, why it is worth using them, and how to configure them in the most popular AI tools. I will also present the most interesting servers for developers and non-technical users.

2. What are MCP servers?



Let's start with the basics. Model Context Protocol is an open communication protocol that standardizes the way AI models connect to external data resources and tools. We can compare it to USB-C for artificial intelligence — a universal connector that works everywhere.

2.1. The N×M problem that MCP solves



Before the appearance of MCP, every AI integration with an external tool required writing dedicated code. If you had N AI models and M tools, you had to create N×M different integrations. This was a nightmare for both developers and companies wanting to use AI in their systems.

MCP solves this problem by creating a universal standard. Now it is enough to implement an MCP server once for a given tool, and all AI models supporting this protocol can use it. The N×M problem turns into an N+M problem — a dramatic reduction in complexity.

2.2. How does it work in practice?



MCP servers are lightweight programs that act as bridges between the AI model and an external data source or tool. They can be run locally on your computer (for access to files, databases, or applications) or remotely in the cloud (for internet services).

The protocol defines three main components:
  • Resources: data that the server makes available to the AI model — these can be files, database entries, or documentation

  • Tools: functions that the AI model can call — e.g., saving a file, sending a message, or executing a SQL query

  • Prompts: predefined prompt templates that facilitate work with a given tool


Communication takes place via JSON-RPC 2.0, which makes the protocol lightweight and universal. MCP was designed with security in mind — it supports OAuth 2.0, authentication, and granular permission control.

3. Why use MCP servers?



The benefits of using MCP are huge, both for individual users and for entire teams.

3.1. Elimination of hallucinations and access to up-to-date data



One of the biggest problems of language models are hallucinations — situations in which AI invents information. MCP radically reduces this phenomenon by giving models access to real, current data. Instead of relying on knowledge from the training period, the model can reach directly into your database, documentation, or API.

3.2. Automation of complex workflows



With MCP, your AI assistant ceases to be a passive chatbot and becomes an active agent that can perform multi-stage tasks. Examples:
  • Server log analysis, problem identification, and automatic creation of a ticket in the ticketing system

  • Retrieving data from the database, generating a report, and sending it via Slack

  • Searching the code repository, finding security bugs, and creating pull requests with fixes


3.3. Standardization and reduction of development costs



Thanks to the open standard, organizations can build integrations once and use them with various AI models. This means:
  • Lower development costs

  • Easier system maintenance

  • Possibility to change AI provider without rewriting integrations

  • Faster implementation of new functionalities


3.4. Security and access control



MCP was designed with corporate security in mind. The protocol supports:
  • Granular permission control — you can specify which resources the model has access to

  • OAuth 2.0 and other authentication mechanisms

  • Resource Indicators (RFC 8707), which prevent access token abuse

  • Audit of all actions performed by AI


However, it is worth remembering that MCP security is still evolving. In 2025, researchers detected several vulnerabilities, including prompt injection problems and potential data exfiltration. It is important to use only trusted MCP servers and regularly update their versions.

4. How to add MCP servers to LLMs?



Now let's move on to practice — I will show you how to configure MCP servers in the most popular AI tools.

4.1. Claude Desktop — simplest configuration



Claude Desktop currently offers the most polished support for MCP. Since September 2025, Desktop Extensions have been introduced — .mcpb packages that you install with one click, without the need to edit configuration files.

Method 1: Installation via extensions catalog (recommended)

For users of paid plans (Pro, Max, Team, Enterprise):
  1. Open Claude Desktop and go to Settings → Extensions

  2. Click Browse extensions to browse the official catalog

  3. Select the extension you are interested in and click Install

  4. Configure the required settings (e.g., API keys) in the friendly interface

  5. The extension is immediately available in conversations


Method 2: Manual configuration (for advanced users)

For users of the free plan or their own MCP servers:
  1. Open Claude Desktop and go to Settings → Developer

  2. Click Edit Config — the claude_desktop_config.json file will open

  3. Add the MCP server configuration in JSON format:


json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/username/Desktop",
        "/Users/username/Downloads"
      ]
    },
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "twoj-token-github"
      }
    }
  }
}


  1. Save the file and restart Claude Desktop

  2. After restarting, you will see a hammer icon in the bottom right corner — click to see available tools


Note: You need Node.js installed on your computer. Check the installation by running node --version in the terminal.

4.2. ChatGPT — Developer Mode with full MCP support



OpenAI added full support for MCP in September 2025 via Developer Mode. It is a powerful, but — as OpenAI itself warns — "powerful but dangerous" functionality.

How to enable Developer Mode:

  1. Log in to ChatGPT (Plus or Pro plan required)

  2. Go to Settings → Connectors → Advanced

  3. Enable Developer Mode

  4. Click Create to add a new MCP connector


Remote MCP server configuration:

json
{
  "name": "Mój serwer MCP",
  "url": "https://twoj-serwer-mcp.com/sse",
  "transport": "sse",
  "auth": {
    "type": "oauth",
    "client_id": "twoj-client-id",
    "client_secret": "twoj-secret"
  }
}


Important information about Developer Mode:
  • ChatGPT can perform write operations — update data, send messages, modify files

  • Always verify actions before executing them (confirmation is required by default)

  • Does not support local servers — you need a tunnel like ngrok for local applications

  • Available only in the browser, not in the mobile app

  • Supports SSE and Streaming HTTP protocols


Security: OpenAI warns that misconfigured connectors or prompt injection attacks can lead to data loss or unintended actions. Use only trusted MCP servers and verify all write operations.

4.3. Other tools supporting MCP



The MCP ecosystem is growing fast. Here are other popular tools with support for this protocol:

Cursor IDE

Cursor, a code editor based on VS Code with built-in AI, supports MCP natively. Configuration in the .cursor/mcp.json file:

json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "twoj-token"
      }
    }
  }
}


Cline (formerly Claude Code)

Console tool for Linux, macOS, and Windows (via WSL). Installation via CLI:

bash
npm install -g @anthropic-ai/cline
cline mcp add github https://github.com/mcp-server-url


Windsurf and Zed

Modern code editors with built-in support for MCP from day one. Do not require installation of additional extensions.

JetBrains IDEs

JetBrains is working on MCP integration in IntelliJ IDEA, PyCharm, and WebStorm. The MCP server runs locally in the IDE and provides context of the whole project for AI suggestions.

5. Best MCP servers for developers



Let's get down to specifics — which MCP servers really change the way developers work?

5.1. GitHub MCP — code management without leaving the editor



Official server from GitHub — one of the most popular MCP servers. Allows for:
  • Browsing repositories, commits, and pull requests

  • Creating and managing issues

  • Commenting and code review

  • Searching in code and history


Why it's worth it: Stay in your flow. Instead of switching between the IDE and the browser, just ask the question: "What issues are assigned to me in project X?" or "Create a pull request with my changes in the feature/auth branch".

Documentation and resources:


Installation:

json
{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "ghp_your_token_here"
      }
    }
  }
}


5.2. Filesystem MCP — intelligent file management



Fundamental server from Anthropic that gives AI access to your file system with controlled permissions.

Capabilities:
  • Reading and writing files

  • Creating and deleting directories

  • Searching for files by patterns

  • Analyzing project structure


Usage example: "Find all TypeScript files in the project that contain unused imports and remove them" — AI will scan the project, find problems, and automatically fix them.

Documentation:


Security: You specify the exact paths the AI has access to. It cannot go outside these directories.

5.3. Docker MCP — containerization from chat



For developers working with Docker, this server is a game-changer.

Features:
  • Listing and managing containers

  • Starting and stopping services

  • Executing commands in containers

  • Inspecting logs and application status


Real-world scenario: "Restart database and backend containers, show logs from the last 5 minutes" — all without leaving the code editor.

Resources:


5.4. Sequential Thinking MCP — AI that thinks out loud



This is not a typical data access server — Sequential Thinking adds the ability of reflective problem solving to AI.

How it works: The model divides complex problems into a sequence of thoughts, showing its reasoning process step by step. Each step can question or develop previous conclusions.

Ideal for:
  • Debugging complicated errors

  • Designing system architecture

  • Solving algorithmic puzzles

  • Application performance analysis


Documentation:


5.5. PostgreSQL MCP — databases in natural language



⚠️ Note: The official PostgreSQL server from Anthropic was archived in May 2025 due to detected security vulnerabilities (SQL injection). We do not recommend using the version @modelcontextprotocol/server-postgres.

Safe alternative:


SQL queries have never been so simple — for safe implementations.

What it offers:
  • Executing SQL queries via natural language

  • Database schema analysis

  • Query optimization

  • Migrations and data management


Example: "Show me the 10 most active users from the last month and their average order value" — AI will write and execute the appropriate query itself.

5.6. Context7 MCP — up-to-date documentation always at hand



One of the biggest problems of AI in programming is outdated knowledge about libraries and frameworks. Context7 solves this problem.

Functionality:
  • Access to the latest documentation of thousands of libraries

  • Semantic search in docs

  • Code examples and best practices

  • Support for the most popular languages and frameworks


Example: "How to use React Hooks in version 18.3?" — AI will reach for the latest documentation instead of relying on outdated knowledge.

Documentation:


5.7. AWS and Azure MCP — cloud under control



Microsoft and AWS have released official MCP servers for their cloud platforms.

AWS MCP Server offers:
  • Management of Lambda, ECS, EKS resources

  • Access to documentation and best practices

  • Billing and performance metrics

  • Deployment and application scaling


AWS Documentation:


Azure MCP Server provides:
  • Integration with Azure DevOps

  • Management of Cosmos DB and Storage

  • Access to Azure CLI

  • Monitoring and logs


Azure Documentation:


5.8. Playwright MCP — UI test automation



Official server from Microsoft. End-to-end tests written in natural language? It is now possible.

Capabilities:
  • Automatic form filling

  • Testing user interactions

  • Screenshots and session recording

  • Cross-browser testing


Demo: "Test the registration form — fill all fields with correct data, submit, and verify that the user is redirected to the dashboard" — Playwright will execute the entire sequence automatically.

Documentation:


5.9. Microsoft Learn Docs MCP — always up-to-date .NET documentation



Official server from Microsoft for .NET developers — this is a must-have. The server provides access to official Microsoft Learn documentation in real-time.

Why it's important: AI often doesn't know the latest C# or .NET features. This server solves the problem by giving access to current API references, tutorials, and best practices.

Documentation:


5.10. Git MCP — version control from AI



Git management without memorizing commands.

Features:
  • Committing changes with intelligent messages

  • Branch management

  • History and diff between versions

  • Resolving merge conflicts


Note: The official GitHub MCP Server (described earlier) provides the most complete Git support, especially for GitHub workflow. For repositories outside GitHub, check:


6. Best MCP servers for non-technical users



MCP is not just a toy for programmers — non-technical people can also significantly improve their work.

6.1. Google Drive MCP — intelligent document management



Your Google Drive becomes fully searchable and editable by AI.

What you can do:
  • Searching documents by content

  • Creating and editing files

  • Organizing folders

  • Sharing and managing permissions


Example: "Find all presentations from Q4 2024 related to marketing and move them to the Archives folder".

Status: The official server from Anthropic was archived in May 2025. Community implementations are available for Google Drive and the entire Google Workspace — details later in the article.

6.2. Slack MCP — team communication without switching windows



Stay up to date with Slack without constantly checking the app.

Capabilities:
  • Reading unread messages

  • Sending messages and replying in threads

  • Searching in conversation history

  • Managing channels and reminders


Use case: You are working in a code editor and want to be up to date with the team: "Check if there were any important messages on #devops in the last hour".

Status: The official server from Anthropic was archived in May 2025. Most popular community implementation:


6.3. Notion MCP — knowledge and task management



Official server from Notion. Notion is a popular tool for project management and note-taking. With MCP you can manage it via AI.

Features:
  • Retrieving and updating pages

  • Creating tasks and projects

  • Searching in notes

  • Managing Notion databases


Example: "Show me all Notion tasks marked as 'High Priority' and remind me about them tomorrow morning".

Documentation:


6.4. Gmail and Google Calendar MCP — productivity to the max



Manage emails and calendar without switching tabs.

Gmail MCP:
  • Searching and filtering emails

  • Sending messages with intelligent replies

  • Managing labels and folders

  • Inbox analysis


Google Calendar MCP:
  • Adding and editing events

  • Checking availability

  • Reminders and summaries

  • Integration with other tools


Workflow: "Check my calendar for this week and send an email to the team with a summary of upcoming meetings".

Community implementations:


6.5. Brave Search MCP — private web search



Official server from Brave. Give your AI access to the internet while maintaining privacy.

Why Brave:
  • Does not track users

  • Fast and accurate results

  • Dedicated API for developers

  • 2000 free queries per month


Features:
  • General and local search

  • Filtering results by date, safety

  • Support for technical queries


Documentation:


6.6. Figma MCP — design meets development



Official server from Figma (Open Beta). For designers and developers working with Figma.

What it offers:
  • Searching projects and components

  • Inspecting styles and properties

  • Exporting assets

  • Generating code from design


Example: "Show me all colors used in the Dashboard project and generate CSS variables".

Documentation:


6.7. Excel and Google Sheets MCP — data analytics simplified



Analyze and modify spreadsheets in natural language.

Capabilities:
  • Data analysis and creating charts

  • Automatic formulas and functions

  • Data cleaning and transformation

  • Generating reports


Use case: "Analyze sales from this sheet, group by months, and create a trend chart".

Community implementations:


6.8. Puppeteer MCP — browser automation



Web scraping and automation without writing code.

Features:
  • Automatic form filling

  • Screenshots and PDFs from pages

  • Navigation and clicking

  • Data extraction from pages


Example: "Go to this page, fill out the contact form, and send a message".

Status: The official server from Anthropic was archived in May 2025. We recommend using the official Playwright MCP Server from Microsoft (described earlier), which offers better capabilities and is actively developed.

6.9. YouTube MCP — channel management and analytics



For YouTube content creators.

Capabilities:
  • Channel statistics and analytics

  • Managing videos and playlists

  • Comments and interactions

  • SEO optimization of videos


Community implementations:


6.10. Google Workspace MCP — everything in one place



If you need a comprehensive solution for the entire Google ecosystem, there is one server that combines it all.

taylorwilsdon/google_workspace_mcp — the most complete integration:


What it includes:
  • Gmail (sending, reading, drafts, labels)

  • Google Drive (files, folders, sharing)

  • Google Calendar (events, availability, invitations)

  • Google Sheets (read/write, formulas)

  • Google Docs (creation, editing)

  • Google Slides (presentations)

  • Google Forms, Tasks, Chat, Custom Search


Why this solution:
  • Production-ready with OAuth 2.1

  • One installation for all Google services

  • One-click setup for Claude Desktop

  • Actively developed and maintained


7. Docker MCP Toolkit — managing MCP servers in one place



Docker Desktop has introduced MCP Toolkit — a tool that significantly facilitates managing local MCP servers.

What it offers:
  • MCP server catalog with one-click installation

  • Automatic management of dependencies and containers

  • Central configuration for all AI tools

  • Easy updates and removal of servers


How to use:
  1. Install Docker Desktop with MCP Toolkit

  2. Go to the MCP tab in Docker Desktop

  3. Browse the catalog and install servers with one click

  4. Configure connection with Claude Desktop or other tools


This is an excellent solution for people who want to experiment with multiple MCP servers without manually configuring each of them.

8. Security and best practices



8.1. Current MCP security challenges



Although MCP is an open-source protocol with a well-thought-out architecture, security researchers have identified several significant problems:

Prompt Injection

Malicious MCP servers may try to manipulate prompts sent to the AI model, forcing it to perform unwanted actions. OpenAI calls this "powerful but dangerous".

Token Misuse

Problem with improper use of access tokens. That is why since June 2025 MCP requires the implementation of Resource Indicators (RFC 8707), which limit the scope of tokens to specific servers.

Lookalike Tools

Possibility of one tool impersonating another, which may lead to data leakage.

NeighborJack

Vulnerability detected in July 2025 that could expose MCP servers in the local network, giving attackers access to the entire host.

8.2. How to protect yourself against threats?



  1. Use only trusted servers: Install servers from official GitHub repositories, Anthropic Registry, or from proven providers

  2. Verify permissions: Before running the server, check what resources it will have access to

  3. Update regularly: Keep MCP servers and AI tools on the latest versions

  4. Do not share publicly: Local servers should remain local — do not expose them to the internet without appropriate security measures

  5. Use OAuth: When possible, use OAuth instead of static tokens

  6. Audit logs: Monitor what actions AI performs via MCP servers

  7. Environment isolation: Consider running MCP servers in Docker containers for an additional layer of security


9. The Future of Model Context Protocol



MCP is developing at a breakneck pace. The roadmap for November 2025 shows exciting directions of development:

9.1. Asynchronous operations



Currently, MCP is mainly synchronous. Soon, servers will be able to run long-running background tasks, informing the model about their status.

9.2. MCP Registry in production



Public MCP server registry that will facilitate discovering and installing new integrations. Currently in preview, planned GA in the coming months.

9.3. Automatic server detection



Servers will be able to announce their capabilities via .well-known URLs, which will enable automatic cataloging by the registry.

9.4. Horizontal scaling



Better support for deploying MCP servers in production environments with load balancing and session management.

9.5. Support in Gemini



Demis Hassabis, CEO of Google DeepMind, confirmed in April 2025 that upcoming versions of Gemini will support MCP natively.

10. Key resources and documentation



If you want to explore the MCP topic, here are the most important sources:

Official MCP documentation:


GitHub Repositories:


Articles and announcements:


Community server catalogs:


11. Summary



Model Context Protocol is not just another technology — it is a fundamental change in how we interact with artificial intelligence. We have moved from isolated chatbots to integrated agents that can act in the real world.

Key takeaways:

  1. MCP is an industry standard: Adoption by OpenAI, Google DeepMind, Microsoft, and Anthropic confirms that this is the future of AI integration

  2. Something for everyone: You don't have to be a programmer to use MCP — servers for Gmail, Drive, or Notion are equally useful

  3. Security matters: Use only trusted servers and monitor permissions

  4. The ecosystem is growing: Hundreds of new MCP servers are created every month — you will find the right tool for every use case

  5. Claude Desktop leads: Currently the most polished MCP integration, with Desktop Extensions facilitating installation

  6. ChatGPT is catching up: Developer Mode gives full write capabilities, though it requires caution

  7. This is just the beginning: The roadmap shows that MCP will be even more powerful with async, registry, and better scaling


Will MCP replace programmers? Of course not. Just as a hammer didn't replace builders, and Excel didn't replace accountants, MCP won't replace developers. But it will change the way we work — allowing us to focus on solving problems instead of writing boilerplate and managing integrations.

© 2025 Jakub Tutka. Built with Nuxt & Vuetify.