MCP (Model Context Protocol) is an open standard that lets AI applications like Claude or ChatGPT connect to outside data and tools, such as your files, databases, calendar or project tool, in one consistent way. The official documentation compares it to a USB-C port for AI applications: instead of a custom connection for every pair of tools, there is one plug that fits all of them. For a team, MCP means you can let the AI assistant you already use read and update your company's work, with permissions you control.
Where MCP comes from
Anthropic introduced MCP as an open standard on November 25, 2024. The problem it described: AI models are cut off from data stuck in silos, and every new data source needed its own custom integration (Anthropic, 2024). The first release included the specification, software development kits (SDKs), local MCP support in the Claude desktop apps and open-source MCP servers for tools like Google Drive, Slack, GitHub and Postgres.
On December 9, 2025, Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. At that point Anthropic reported more than 10,000 active public MCP servers and over 97 million monthly SDK downloads across Python and TypeScript, and named ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code among the products that had adopted it (Anthropic, 2025).
In short: MCP is not one company's feature anymore. It is a shared standard.
How MCP works, in plain terms
There are three roles (Model Context Protocol, 2026):
- Host: the AI application you use, for example Claude Desktop, ChatGPT or an IDE.
- Client: a component inside the host that keeps one connection to one server.
- Server: a program that offers data and actions from a specific system, for example your project tool or your file storage.
The host creates one client for each server it connects to. A server can run locally on your computer or remotely on the vendor's side. Remote servers usually use a web-based transport (Streamable HTTP), and the specification recommends OAuth for authentication.
A server can offer three kinds of things:
| Primitive | What it is | Example in a project tool |
|---|---|---|
| Tools | Actions the AI can take | Create a task, mark it done, post a comment |
| Resources | Data the AI can read for context | A project's tasks, a knowledge page |
| Prompts | Reusable templates for common jobs | "Prepare a weekly status update" |
The AI application asks the server which tools it offers, and the model decides when to call them during a conversation. Servers can also ask the user for more information or for confirmation before an action, a feature the specification calls elicitation.
A simple example
Say your team's tasks live in a project tool with an MCP server, and you use Claude.
- You connect Claude to the project tool's MCP server and sign in. Claude now sees the tools the server offers, such as "list my tasks" and "create task."
- You ask: "What's overdue on the website project, and draft a message to the owners?"
- Claude calls the server's tool to read the project's tasks, finds the overdue ones and writes the message.
- If you say "create a follow-up task for each," Claude calls the create tool. Depending on the server and your settings, you confirm first.
Without MCP, you would copy and paste the task list into the chat, and nothing you decided would flow back.
Why MCP matters for businesses
- Use the AI you already like. Your team can work in Claude or ChatGPT and still use the company's real data.
- Build once, connect everywhere. A vendor builds one MCP server instead of separate integrations for each AI app.
- AI stops starting from zero. The assistant can read your goals, tasks and knowledge instead of guessing. See How to give ChatGPT or Claude access to company knowledge.
- Agents become practical. AI agents need tools to act; MCP is a standard way to give them those tools. See What is an AI agent for business?
What to watch out for
MCP makes connections easy, so be deliberate about what you connect:
- Permissions. Check what a server lets the AI read and change. Prefer servers that respect your existing user permissions and let you limit actions to read-only.
- Confirmation for actions. For anything that changes data or sends messages, require a confirmation.
- Trusted servers only. Install servers from the vendor or from sources you trust. A server is code that runs with access to your data.
- Visibility. Choose tools that show what the AI did through MCP, so actions can be checked and undone.
MCP vs. APIs vs. plugins
| Regular API | App-specific plugin | MCP | |
|---|---|---|---|
| Built for | Developers writing code | One AI app | Any MCP-capable AI app |
| AI can discover actions itself | No | Within that app | Yes, the server lists its tools |
| Reuse across AI apps | Needs custom work | No | Yes |
MCP usually sits on top of a product's existing API. It does not replace the API; it describes it in a way AI applications understand.
How Orakis does it
Orakis offers MCP so that Claude, ChatGPT or any other MCP-capable assistant can work directly with an Orakis workspace: its goals, tasks and company knowledge. That means you can ask your own assistant what is late, or have it create work, and the result lands in the same workspace your team uses. Orakis also has its own built-in agent, Alfred, so MCP is an option, not a requirement. Leads decide how much Orakis may approve on its own, and everything it approves is visible. See Project management tool that works with Claude and ChatGPT for more.
FAQ
Is MCP only for Claude? No. Anthropic created it, but it is now governed under the Linux Foundation, and ChatGPT, Gemini, Microsoft Copilot, Cursor and Visual Studio Code are among the products that support it.
Do I need to be a developer to use MCP? To use it, no. In many AI apps you add a connector or a server address and sign in. Building your own MCP server does take development work.
Is MCP secure? The protocol supports standard authentication such as OAuth, and servers can ask for confirmation before actions. Security still depends on the server you connect and the permissions you grant, so connect only trusted servers and limit what they may change.