By Fahad Murtaza · Published
MCP Server Examples: Real Servers, Tools and Calls
An MCP server is a program that offers a set of capabilities to an AI application through the Model Context Protocol. The quickest way to understand the idea is to look at real ones. This page walks through the main kinds, what each exposes, and one worked example on a WordPress site. If the term is new, read what an MCP server is first.
What a server can expose
The protocol defines three building blocks:
- Tools are actions the model can call, each with a name, a description and an input schema. This is where nearly all the value is.
- Resources are read-only data the client can attach as context, such as a file or a record.
- Prompts are reusable templates a user can pick, often shown as slash commands.
A client asks for the lists at runtime (tools/list, resources/list, prompts/list), so a new server needs no custom code in the client. That is the main difference from a REST API, covered in MCP vs API.
Examples by category
These are the kinds of servers you will meet most often. Names of specific products change quickly, so the categories matter more than any one listing.
Filesystem servers
A filesystem server lets a model read, search and write files inside folders you allow. Typical tools are read_file, write_file, list_directory and search_files. It runs locally over stdio. The main risk is scope: a server pointed at your home folder can read far more than the task needs, so give it one project directory.
Database servers
A database server exposes a schema and a query tool. Good ones are read-only by default, because a model that can run arbitrary writes can also drop a table. Look for a separate, explicit write mode, and connect with a database user that has only the privileges the task needs.
GitHub and developer servers
These wrap a hosted API: list issues, open a pull request, read a diff, comment on a review. They are usually remote or run locally with a token. The token’s scopes are the permission boundary, so create a token limited to the repositories you want the agent to touch.
Browser automation servers
A browser server drives a real browser: navigate, click, fill a form, take a screenshot, read the page text. It is useful for testing a front end or checking what a visitor sees. It is also the broadest kind of access, since the agent acts as whoever is logged in. Use a clean browser profile.
Documentation search servers
These index a body of documentation and expose a search tool, sometimes with resources for each page. They are read-only and low risk, and they cut down on a model inventing an API that does not exist. Many are hosted remotely and need no credentials.
CMS and site servers
A content management system is a natural fit, because the work is a long list of named operations: create a post, change a menu, update a product. That is the case for WordPress, shown next.
Reference servers you can run today
The Model Context Protocol project keeps a set of reference servers in the modelcontextprotocol/servers repository. They are small, readable and a good way to see the pattern. At the time of writing the active ones are:
| Server | What it does | How it is started |
|---|---|---|
| Filesystem | Read, write and search files in folders you list | npx -y @modelcontextprotocol/server-filesystem /path/to/folder |
| Git | Read, search and change a Git repository | uvx mcp-server-git |
| Fetch | Fetch a web page and convert it for the model | uvx mcp-server-fetch |
| Memory | A small knowledge graph the model keeps between chats | npx -y @modelcontextprotocol/server-memory |
| Everything | A test server that exercises prompts, resources and tools | npx @modelcontextprotocol/server-everything |
| Sequential Thinking and Time | A thinking scratchpad, and time zone conversion | See the repository |
Several older reference servers, such as GitHub, PostgreSQL, Slack and Puppeteer, have moved to an archive repository, and some now have vendor-maintained successors. Check a server’s repository for its current status before you rely on it.
Two things are worth doing with these. Run the Filesystem server through MCP Inspector and read the tools/list reply: you will see the same structure of name, description and input schema that every server uses, including the WordPress one below. And read how little code the Fetch or Time server needs, because that is the honest size of a basic MCP server. The build guide takes you through writing one.
Worked example: a WordPress site
WP MCP is a plugin that turns a WordPress site into a remote MCP server. It registers its tools through the WordPress Abilities API and serves them at https://your-site.com/wp-json/mcp/wpmcp-server. It registers up to 407 tools across 61 domains: 274 in the free plugin and 133 in Pro. The count on a given site depends on which integrations are active, and list-tool-catalog shows exactly what that site has.
A short session, with the calls an agent makes behind each request:
You: List my last five draft posts.
{ "name": "list-posts", "arguments": { "status": "draft", "per_page": 5 } }
You: Change the title of post 482 to "Spring Menu 2027".
{ "name": "update-post", "arguments": { "id": 482, "title": "Spring Menu 2027" } }
The write returns an operation_id, because WP MCP stores a snapshot before every supported write. Undoing it is another tool call:
{ "name": "list-operations", "arguments": {} }
{ "name": "rollback-operation", "arguments": { "operation_id": "<id from the list>" } }
rollback-session unwinds everything an agent did in one session. The same undo is available as a Restore button in wp-admin. A few kinds of change cannot be snapshotted, such as package updates and raw database row writes, and the README lists them under “What cannot be undone” instead of claiming an undo that does not exist.
What this server exposes, in terms of the three building blocks:
- Tools: posts, pages, media, menus, settings, users, plugins, themes and the safety tools above, plus builder and integration tools. See the WordPress MCP server page for the full picture.
- Prompts and resources: the plugin serves its bundled skills as MCP prompts, and offers
wpmcp://site/contextandwpmcp://skillsas resources, for clients that show those natively.
For a store, the same endpoint covers products and orders: see MCP for WooCommerce and the WooCommerce integration.
How to judge an MCP server
Whatever the category, ask the same six questions.
- What can it change? List the write tools. A server with no writes is the easiest to trust.
- How does it authenticate? Prefer a scoped, revocable credential. WP MCP accepts a WordPress application password, and OAuth 2.1 when the site owner switches it on. See MCP OAuth explained.
- Can a mistake be undone? Few servers say. Ask whether writes are recorded and restorable, and which ones are not.
- Is there an audit trail? You want to know which client did what and when.
- How big is the tool list? Every definition costs tokens on each connection. A compact mode or a way to filter tools helps.
- Is it maintained? Check the release dates and the open issues, not only the star count.
The security side is covered in MCP security for WordPress.
Try one on a real site
The fastest way to see the difference between a list of tools and a working server is to connect one. For a WordPress site, install WP MCP, create an application password and follow Add an MCP server to Claude Code, or use the steps for Claude or Cursor. The free plugin includes the safety engine and Gutenberg tools; pricing covers the Pro tier for page builders and the rest. Building your own is possible too: see how to build an MCP server.
To see how tools, resources and prompts differ, read MCP tools explained and MCP resources and prompts.
Common questions
What are some examples of MCP servers?
Common examples are a filesystem server that reads and writes files, a database server that runs queries, a GitHub server that manages issues and pull requests, a browser automation server, a documentation search server, and a CMS server such as WP MCP that lets an AI client manage a WordPress site.
What does an MCP server expose?
Up to three kinds of things: tools the model can call, resources the client can read as context, and prompts that act as reusable templates. Most servers expose tools only. A client learns what is available by asking the server for its lists at runtime.
Do MCP servers run locally or remotely?
Both. A local server runs as a process on your machine and talks over standard input and output. A remote server runs at a URL and talks over HTTP. A WordPress site is naturally a remote server, because the site already lives at a URL.
How many tools should an MCP server have?
Enough to cover the job, and no more than the client can handle. Every tool definition costs tokens on each connection, and some clients cap the count. WP MCP registers up to 407 tools and offers a compact mode that advertises five.
How do I check that an MCP server is safe to use?
Look at what it can change, how it authenticates, whether writes can be undone, and whether it logs what the agent did. Start with a read-only credential, use a dedicated user, and test on a staging copy before connecting a live site.