What is an MCP server? The Model Context Protocol, explained
MCP is the open standard that lets AI apps like Claude and ChatGPT plug into your tools and data. An MCP server is the plug on the tool’s side. Here is how it works, in plain words, with sources.
What is an MCP server?
An MCP server is a small program that lets an AI app use an outside tool or data source through the Model Context Protocol, an open standard. It tells the AI what it can do, such as search files or read a calendar, and carries out those actions when the AI asks. It runs locally or online.
- MCP stands for
- Model Context Protocol
- Introduced by
- Anthropic, as an open standard, on November 25, 2024
- Run by
- Agentic AI Foundation, under the Linux Foundation, since December 2025
- An MCP server offers
- Tools (actions), resources (data) and prompts (templates)
- Adoption
- 10,000+ active public servers by December 2025; used in ChatGPT, Cursor, Gemini, Copilot, VS Code
What does MCP stand for in AI?
In AI, MCP stands for Model Context Protocol. The official documentation describes it as an open-source standard for connecting AI applications to external systems, such as files, databases, search engines and other apps.
Anthropic, the company behind Claude, introduced MCP on November 25, 2024. Its reasoning: even the best models were cut off from data, and every new data source needed its own custom connection. MCP replaces those one-off connections with one shared standard. Anthropic launched it with ready-made servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer.
On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. So MCP is no longer one company’s project.
What is an MCP server, exactly?
An MCP server is the program on the tool’s side that offers context and actions to AI apps. The MCP architecture docs name three parts:
| Part | What it is | Example |
|---|---|---|
| Host | The AI app you use, which manages the connections | Claude Code, Claude Desktop, Visual Studio Code |
| Client | A connector inside the host, one per server | The link VS Code opens to a Sentry server |
| Server | The program that offers tools and data to the AI | A filesystem server, a database server, Sentry’s server |
A server can offer three kinds of things, which the docs call primitives:
- Tools: actions the AI can take, such as querying a database, calling an API or editing a file.
- Resources: data the AI can read, such as file contents, database records or API responses.
- Prompts: reusable templates that help the AI use the server well, such as example questions.
Servers can be local or remote. A local server runs on your own computer and talks to the app directly; the docs’ example is Claude Desktop starting a filesystem server. A remote server runs online over the web, like Sentry’s official MCP server, and can serve many users at once.
How does MCP work?
MCP works as a simple conversation between the AI app and the server: the app asks what the server can do, the AI picks a tool when it needs one, and the server does the work and sends back the result.
- You connect a server to your AI app, for example a calendar server.
- The app asks the server what it supports and gets a list of tools, each with a name, a description and the inputs it needs.
- You ask the AI, “Move my 3pm call to Friday.”
- The AI sees a matching tool and asks the app to call it with the right inputs.
- The server makes the change in your calendar and sends back the result.
- The AI tells you it is done, or asks you to confirm first if the app requires it.
Under the hood, messages use JSON-RPC 2.0, a common format for one program to ask another to do something. Servers can also tell the app when their list of tools changes, so the AI always knows what is available. None of this is something a non-developer has to touch: you add a server in your app’s settings, and the app handles the rest.
What is MCP and what are its advantages?
The main advantage of MCP is “build once, use everywhere”: a tool that offers one MCP server works with every AI app that supports MCP, instead of needing a separate integration for each.
| Without MCP | With MCP | |
|---|---|---|
| Connecting 5 AI apps to 5 tools | Up to 25 custom integrations | Each tool ships one server; each app supports MCP once |
| Switching AI app | Rebuild your integrations | Reconnect the same servers |
| Who builds the connector | Every AI app, for every tool | Usually the tool’s maker, once |
| What the AI can do | Mostly talk | Read live data and take actions |
- For developers: the docs say MCP reduces development time and complexity when building or integrating with an AI app or agent.
- For AI apps: access to a wide ecosystem of data sources, tools and apps.
- For people using AI: assistants that can use your data and take actions for you, when needed.
Adoption is broad. By December 2025, Anthropic counted more than 10,000 active public MCP servers and over 97 million monthly SDK downloads across Python and TypeScript, and listed ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code among the products that support it.
MCP vs API: what is the difference?
An API is how one program talks to a specific service; MCP is a standard way for AI apps to discover and use many services. MCP does not replace APIs. Many MCP servers are thin layers on top of a company’s existing API that describe it in a way an AI can understand and use.
| API | MCP | |
|---|---|---|
| Built for | Developers writing code | AI apps choosing tools on their own |
| Format | Different for every service | The same for every server |
| Discovery | A developer reads the docs | The AI app asks the server what it offers |
| Typical use | Your app calls Stripe, Google Maps or OpenAI | Claude or ChatGPT uses your files, calendar or tickets |
Is MCP safe? The risks to know
MCP is as safe as the servers you connect and the permissions you give them. A server can act on your data, so a bad or careless one can do damage. The official security best practices spell out the main risks.
- Local servers run code on your machine with the same rights as your AI app. The docs warn they can lead to stolen data or data loss if they come from an untrusted source.
- Apps that offer one-click setup must show the exact command and ask for your approval before running a new local server.
- Servers must not accept access tokens that were issued for other services and pass them along.
- Ask for as few permissions as possible; a stolen token with broad access does far more harm.
Do you need MCP to build an AI app?
No. MCP matters when you want an AI assistant to reach your tools, or you want your product to be usable from inside AI assistants. If you are building an AI app for customers, the bigger jobs are the app, payments, AI costs and support. Our guide on how to build an AI app covers those, and AI agents vs agentic AI explains the agents MCP often feeds.
Developers who build from coding agents use MCP every day to give those agents access to databases, docs and deploy tools. If that is you, compare Codex vs Claude Code. If you would rather describe the app and not wire anything, an AI app builder does the setup for you.
Use Starty in its own chat, or from Claude Code and Codex
Starty is our product. It builds and runs AI apps, and the Starty agent works in Starty’s own chat or through MCP from coding tools such as Claude Code and Codex.
- Developers can reach Starty from Claude Code or Codex over MCP, and skip building payments, emails and support themselves.
- Non-developers just describe the app; sign-in, Starty Payments, emails, a customer agent, legal pages and analytics come set up in test mode.
- AI features like vision, voice, image, video and agents switch on in one click, with no API keys, metered against your credits.
- Nothing sends, charges, refunds, publishes or deletes without your OK.
Related guides
Where every fact comes from
Official pricing pages, docs, announcements and named studies, checked on November 11, 2026. Where a company doesn’t list something, we say so rather than guess.
- Anthropic: Introducing the Model Context Protocol (November 25, 2024)
- Model Context Protocol: What is MCP?
- Model Context Protocol: Architecture overview
- Model Context Protocol: Security best practices
- Anthropic: Donating MCP and establishing the Agentic AI Foundation (December 9, 2025)
- Sentry: MCP server
Questions people ask
MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic in November 2024, that lets AI apps connect to outside tools and data in one common way.
Anthropic created MCP and released it as an open standard. In December 2025 it donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg.
Yes. The official MCP documentation lists ChatGPT among the AI assistants that support MCP, alongside Claude, and development tools such as Visual Studio Code and Cursor.
No. An API is a service’s own interface for developers. An MCP server describes tools and data in a standard way that any MCP-compatible AI app can discover and use. Many MCP servers are built on top of an existing API.
Yes. The project publishes free SDKs in several programming languages and reference servers to copy from. You need some coding skill; a simple server that offers one or two tools is a common first project.
Only if you trust the source. Local servers run code on your computer with your AI app’s rights. Use servers from the tool’s maker, give them the least access that works, and keep approvals on for risky actions.
Enter your email to join Starty. We’ll email you your invite, and you can start building for free.
No. You describe what you want, or fork an idea, and Starty builds it. You review and approve; you never have to open the code.
No. Starty takes no fee on your revenue. Payouts go to your bank account, and Starty earns from plans and credits.