MCP Server Examples: 12 Real Ways to Extend Claude AI

In this guide

1. Prerequisites: 5 Things You Need
2. No-Code MCP Server Examples
3. Low-Code MCP Server Examples (Requires API Keys)
4. Requires Coding (Node.js or Python)
5. Step-by-Step: Installing the Filesystem Server
6. 3 Common Failures and Fixes
7. Finding More Servers: GitHub Search and the Official List
8. Security and Limitations
9. Frequently Asked Questions

12 MCP server examples that connect Claude to your calendar, Notion workspace, databases, and web search—with setup steps for each and which ones work without coding.

Model Context Protocol (MCP) servers are small programs that bridge Claude to the services you already use. They sit between Claude and another service: Claude sends a request through the server, the server talks to the API, and the result comes back. You configure servers in a JSON file that Claude reads when it starts.

Prerequisites: 5 Things You Need

• Claude Desktop app (available for macOS and Windows from anthropic.com/claude)
• A text editor that can open JSON files (Notepad on Windows or TextEdit on Mac works)
• The GitHub repository link for the MCP server you want to install
• For some servers: an API key from the service you’re connecting (Google, Notion, Slack, etc.)
• For coding-required servers: Node.js 18 or later installed on your machine

The Claude Desktop app must support MCP. Versions released before October 2024 do not. If your config file changes produce no effect, update from the Claude website.

No-Code MCP Server Examples

These servers work with pre-built executables or simple configuration. No terminal commands beyond copy-paste.

1. Filesystem MCP Server

GitHub: modelcontextprotocol/servers (official repository)
What it does: Lets Claude read and write files in folders you specify. Ask Claude to summarize a PDF in your Downloads folder or create a text file with meeting notes.
Setup difficulty: Add four lines to your Claude config file pointing to the folder path.
Business use case: Batch-rename files, extract data from CSVs, generate reports from local documents.
The problem: Point this at your home directory and Claude can read your browser history, SSH keys, and tax returns. Limit it to a single project folder. The official README does not warn about this clearly enough.

2. Brave Search MCP Server

GitHub: modelcontextprotocol/servers
What it does: Claude can search the web through Brave’s API and return current results.
Setup difficulty: Sign up for a free Brave Search API key at brave.com/search/api, paste it into the config.
Business use case: Competitive research, checking if a domain is available, finding recent news about a vendor.
The problem: Brave’s free tier has a monthly query cap (check the Brave API dashboard for your current limit). Hit it mid-month and searches fail silently—Claude just says “I couldn’t find results” with no indication you’ve exhausted your quota.

3. SQLite MCP Server

GitHub: modelcontextprotocol/servers
What it does: Query SQLite databases with natural language. Claude writes the SQL for you.
Setup difficulty: Point the config at your .db file. No SQL knowledge required to ask questions; Claude generates the queries.
Business use case: Small inventory databases, customer lists exported from other tools, analytics data you’ve downloaded.
The problem: Claude can write UPDATE and DELETE queries unless you open the database in read-only mode. The default config does not enforce this. Add ?mode=ro to your database path or risk accidental data changes.

4. Memory MCP Server

GitHub: modelcontextprotocol/servers
What it does: Claude remembers facts across conversations. Tell it your client’s preferred invoice format once; it recalls it next month.
Setup difficulty: Enable in config; no external service needed.
Business use case: Store project preferences, client communication styles, recurring task templates.
The problem: Memories persist until you manually delete them, and there’s no built-in way to list what Claude has stored. You must ask “what do you remember about me?” and trust the answer is complete. The storage file lives in your app data directory as plain JSON.

Low-Code MCP Server Examples (Requires API Keys)

These need an account with a third-party service and an API key, but the server code is ready to run.

5. Google Calendar MCP Server

GitHub: Search mcp-server-google-calendar on GitHub (multiple implementations exist)
What it does: Read your calendar, check availability, create events by telling Claude “schedule a call with Alex next Tuesday at 2pm”.
Setup difficulty: Create a Google Cloud project, enable Calendar API, download OAuth credentials JSON. Paste file path into config.
Business use case: Schedule across team calendars, block time for deep work, generate availability summaries.
The problem: Google’s OAuth token expires after 7 days if you use a test project. You must republish the Cloud project to production or re-authenticate weekly. The community servers vary in quality—some haven’t been updated since early 2024 and fail with current Google API versions.

6. Notion MCP Server

GitHub: makenotion/notion-mcp-server
What it does: Search your Notion workspace, read page content, create new pages. Ask Claude to find all pages tagged “client onboarding”.
Setup difficulty: Generate a Notion integration token from your workspace settings, add it to config.
Business use case: Extract data from meeting notes, compile project status from multiple pages, create templated pages.
The problem: The integration must be manually added to each page or database you want Claude to access—workspace-wide access is not automatic. Many users configure the server, then wonder why Claude says “I don’t see that page.” You must share it with the integration first.

7. Slack MCP Server

GitHub: Search mcp-server-slack on GitHub
What it does: Read channel history, search messages, post replies. Claude can summarize yesterday’s #support channel or find that link someone shared last week.
Setup difficulty: Create a Slack app in your workspace, add read permissions, copy the bot token.
Business use case: Daily digests of key channels, finding decisions buried in threads, drafting announcements.
The problem: Free Slack workspaces limit message history to 90 days. Claude can only search what Slack’s API returns. Also, the bot must be invited to each channel—it cannot read private channels it’s not a member of, even with admin tokens.

8. GitHub MCP Server

GitHub: modelcontextprotocol/servers
What it does: Read repository contents, search code, list issues and pull requests. Ask Claude “show me open bugs labeled ‘critical'”.
Setup difficulty: Generate a GitHub personal access token with repo permissions at github.com/settings/tokens, add to config.
Business use case: Track project progress, review code changes in plain English, find examples of how you solved a problem before.
The problem: Personal access tokens grant access to all repos you can see. There’s no way to limit the token to a single repository. If your config file leaks, someone has read (or write, depending on token scope) access to your entire GitHub account.

Requires Coding (Node.js or Python)

These are powerful but need you to install dependencies and occasionally edit configuration files by hand.

9. Postgres MCP Server

GitHub: modelcontextprotocol/servers
What it does: Connect Claude to a PostgreSQL database. Query production data, generate reports, check table schemas.
Setup difficulty: Install via npm with npm install @modelcontextprotocol/server-postgres, provide a database connection string in the config. Read-only mode is safest.
Business use case: Ad-hoc analytics without writing SQL, explaining schema to new team members, sanity-checking data migrations.
The problem: The connection string includes your database password in plain text in the config file. Use a read-only user account, not your admin credentials. Also, Claude sometimes generates queries that are syntactically correct but slow—no query timeout is enforced by default.

10. Puppeteer MCP Server

GitHub: modelcontextprotocol/servers
What it does: Claude can control a headless browser: take screenshots, fill forms, scrape structured data from pages.
Setup difficulty: Install via npm with npm install @modelcontextprotocol/server-puppeteer. Requires Node.js and Puppeteer dependencies. Works best on macOS/Linux; Windows requires WSL for reliable operation.
Business use case: Monitor competitor pricing, automate repetitive web tasks, test how your site renders.
The problem: Many sites block headless browsers. Puppeteer can be configured to look like a real browser, but the default MCP server config does not do this. Expect CAPTCHAs and access denials on major platforms. The server also has no rate limiting—Claude can spawn dozens of browser instances if you ask it to check 50 URLs.

11. AWS KB Retrieval MCP Server

GitHub: modelcontextprotocol/servers
What it does: Query an AWS knowledge base (if you use Amazon Bedrock). Claude retrieves documents you’ve indexed there.
Setup difficulty: Requires an AWS account, Bedrock setup, IAM credentials configured in your environment. This is for teams already on AWS.
Business use case: Internal documentation search, compliance document retrieval, customer support knowledge bases.
The problem: Bedrock Knowledge Bases cost $0.10 per 1,000 queries (as of AWS pricing in September 2026). A day of heavy Claude use can cost $5–10 in retrieval fees. The server provides no spend alerts. Also, setup requires IAM policies that many corporate AWS accounts lock down—expect to involve your infrastructure team.

12. Custom REST API MCP Server

GitHub: Community examples on GitHub under topic:mcp-server
What it does: Template for connecting Claude to any REST API your business uses—CRM, inventory system, payment gateway.
Setup difficulty: Clone a starter template, edit the endpoints and authentication. Requires reading API documentation and writing JSON schema definitions for each endpoint.
Business use case: Unique to your stack. One small business uses this to let Claude check warehouse stock levels in real time.
The problem: You are writing the glue code between Claude and your API. If the API changes, you must update the server. If your schema definitions are wrong, Claude will send malformed requests. Community templates are often abandoned after a few months—check the last commit date before investing time.

Step-by-Step: Installing the Filesystem Server

The Filesystem server requires no API keys, making it the simplest first install. This walkthrough uses that server as an example. The same config file structure applies to all MCP servers.

Step 1: Locate Your Claude Config File

On macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
On Windows: %APPDATA%\Claude\claude_desktop_config.json

Open it in a text editor. If it doesn’t exist, create a new file with that exact name in that location. The file must be named claude_desktop_config.json, not config.json or claude_config.json.

Step 2: Add the Server Configuration

Paste this JSON structure, replacing /Users/yourname/Documents with the folder you want Claude to access:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/yourname/Documents"
      ]
    }
  }
}

On Windows, the path looks like C:\\Users\\yourname\\Documents. Note the double backslashes—JSON requires them.

Save the file. JSON is strict: every opening brace { needs a closing one }, and commas go between entries but not after the last one. If you already have other servers configured, add a comma after the previous server’s closing brace, then paste the filesystem block.

Step 3: Restart Claude Desktop

Quit Claude completely. On Mac, right-click the dock icon and choose Quit. On Windows, close from the system tray (the app must not be running in the background). Reopen it.

Step 4: Test the Connection

In a new Claude conversation, type: “List the files in my Documents folder.” The expected behaviour: Claude returns a list of filenames from that directory. If it says it cannot access files or does not recognize the request, check the next section.

3 Common Failures and Fixes

Claude Says “I Don’t Have Access to That”

The cause: The config file has a syntax error—a missing comma, an extra comma after the last item, or smart quotes (") instead of straight quotes (").
The fix: Copy your config file contents and paste into a JSON validator (search “json validator” or use jsonlint.com). It will highlight the line with the error. Fix it, save, and restart Claude.

Claude Starts But No MCP Servers Appear

The cause: The config file is in the wrong location, named incorrectly, or saved with a hidden .txt extension.
The fix: On Windows, enable “File name extensions” in File Explorer’s View menu. Check that the file is exactly claude_desktop_config.json, not claude_desktop_config.json.txt. On Mac, check that the file is in ~/Library/Application Support/Claude/, not ~/Documents/.

Server Works Once, Then Stops

The cause: For servers that use npx, the first run downloads the package from npm. If your internet connection drops during this, the download is incomplete.
The fix: Delete the npm cache. On Mac/Linux, run rm -rf ~/.npm in Terminal. On Windows, delete %APPDATA%\npm-cache. Restart Claude with a stable internet connection—it will re-download the package.

Finding More Servers: GitHub Search and the Official List

The official list lives at github.com/modelcontextprotocol/servers. That repository includes 10 maintained servers as of September 2026: filesystem, SQLite, Postgres, GitHub, Brave Search, Google Drive, Puppeteer, Memory, Sentry, and AWS KB Retrieval.

For community-built servers, search GitHub for topic:mcp-server or mcp-server- plus the service name (“mcp-server-airtable”, “mcp-server-stripe”). Check the repository’s last commit date—anything untouched since early 2024 may not work with current Claude versions. Read the README for setup steps: complexity varies from a single config line to multi-step OAuth flows.

The command field in your config tells Claude how to launch the server. npx is for Node.js packages published on npm. python or python3 is for Python scripts. A direct file path like /usr/local/bin/my-server is for compiled binaries. Some servers require environment variables for API keys—these go in an env block in the config.

Security and Limitations

MCP servers start when you begin a Claude conversation and stop when you close the app. They cannot run continuously in the background. They cannot trigger actions on a schedule—you must ask Claude explicitly each time. They do not share state between conversations unless the server writes to a file or database (the Memory server does this; most others do not).

Permissions matter. A filesystem server pointed at your home directory can read your SSH keys, browser cookies, and application passwords. Point it at a single project folder instead. Use read-only database connections for production data—add ?mode=ro to SQLite paths or create a Postgres user with SELECT-only grants. Treat API keys in the config file like passwords: that JSON file is plain text on your disk, readable by any process running as your user.

Some services (Google, Slack, GitHub) let you scope API tokens to specific permissions. Use the minimum required. A GitHub token with full repo access can delete your code if Claude generates a destructive command. A Slack token with write access can post messages as you. The MCP server does what Claude tells it to do—there is no safety layer between Claude’s generated commands and the API call.

Frequently Asked Questions

What are the most useful MCP server examples for beginners?

Start with the Filesystem server (read/write local files) or Brave Search (web search). Both need only a config change and no external accounts beyond the Brave API key. The Memory server is also beginner-friendly—it lets Claude remember facts across chats with zero setup beyond enabling it. These three teach you how MCP works without requiring OAuth flows or coding. Avoid Postgres, Puppeteer, and custom API servers until you are comfortable editing JSON and reading error logs.

Where can I find ready-made MCP servers to use?

The official repository at github.com/modelcontextprotocol/servers has 10 maintained servers as of September 2026, including filesystem, SQLite, Postgres, GitHub, and Brave Search. For community servers, search GitHub for topic:mcp-server or the service name plus “mcp-server” (like “notion mcp-server”). Each repository’s README includes installation steps and required API keys. Check the last commit date—servers abandoned in early 2024 may not work with current Claude versions.

What can MCP servers do for Claude AI?

MCP servers give Claude access to data and services outside the chat window. They let Claude read your calendar, search your company’s Notion wiki, query databases, control a web browser, or call any REST API. The trade-off: you are granting Claude access to real data. A filesystem server can delete files. A database server can run UPDATE queries. A Slack server can post messages as you. Set permissions narrowly and use read-only modes where possible.

Do I need to code to use MCP server examples?

Not for the simplest ones. Filesystem, Brave Search, SQLite, and Memory servers work by editing a JSON config file—no terminal commands beyond copy-paste. Servers that need API keys (Google Calendar, Notion, Slack) require creating a developer account and copying a token, but no coding. Postgres, Puppeteer, and custom API servers require Node.js or Python installed, comfort with command-line tools, and the ability to debug dependency errors. If you have never used npm install or pip install, start with the no-code servers first.

Photo by Pavel Danilyuk on Pexels