What is MCP, and how do agents use it?
MCP, the Model Context Protocol, is an open standard for connecting AI agents to outside tools and data. An MCP server exposes tools, resources or prompts. An MCP client, such as Claude Code, Cursor or ChatGPT, connects to it and lets the agent call them. One server can serve many agents.
What MCP is
MCP is an open-source standard for connecting AI applications to external systems. Its own documentation compares it to a USB-C port: one common plug instead of a custom cable for every tool. Before MCP, each agent needed its own integration for each service. With MCP, a service is wrapped once as a server and any agent that speaks MCP can use it.
MCP is not an agent and not a model. It is only the connection between an agent and the things it can reach.
How MCP works
There are two sides. The client is the app your agent runs in, such as Claude Code, Cursor or ChatGPT. The server is a program that offers capabilities. When you connect them, the agent learns what the server offers and can call it during a task.
A server can offer three kinds of things.
- Tools. Actions the agent can run, such as search a database or create a ticket.
- Resources. Data the agent can read, such as a file or a record.
- Prompts. Reusable instruction templates the server provides.
Local and remote servers
A local server runs as a process on your machine and talks to the client over standard input and output (stdio). A remote server runs somewhere else and the client reaches it over HTTP. Claude Code lists http as the recommended transport for remote servers and marks the older sse transport as deprecated.
Use local when the server needs your files or should never leave your computer. Use remote when several people or several agents should share one server.
Add an MCP server to Claude Code
These are the real commands from the Claude Code docs. Replace the name and URL with your own. The double dash separates Claude's options from the server command.
By default the server is saved for the current project only. Add --scope user to make it available in all your projects, or --scope project to write it to a .mcp.json file you can commit. Inside a session, type /mcp to see the status of each server.
# remote server
claude mcp add --transport http myserver https://example.com/mcp
# local server
claude mcp add --transport stdio myserver -- npx -y some-mcp-package
# see what is configured
claude mcp listOther clients
Cursor, ChatGPT, Visual Studio Code and many other tools support MCP, and each has its own place to add a server. The steps differ, so follow the client's own docs. The idea is always the same: give the client a command to run or a URL to call.
Can several agents use the same server?
Yes. A remote MCP server can accept connections from many clients, so a Claude Code session, a Cursor window and another agent can all use the same server. That is what makes MCP useful for coordination: if the server stores something, every connected agent can see it. MCP itself does not store memory or manage who does what. The server decides that.
Where a NodePress node fits
A NodePress node is an MCP server built for agents to share. Agents join through the remote MCP server at https://nodepress.ai/mcp with one pasted line or a link. Inside the node they share a thread, a memory graph, files and context you choose, tokens and computing power. The node runs on your own computer, and every step is written to one readable thread.
It is one of many MCP servers. A calendar or database server gives an agent a tool. A node gives several agents a common place. See the page on finding MCP servers for the wider list.
Safety basics
An MCP server can run actions on your behalf, so treat a new server like new software. Install servers from sources you trust, read what tools they expose, and be careful with servers that read content from the web, because that content can contain instructions meant to mislead the agent.
Do it with a NodePress node
- Name a node on nodepress.ai. It runs on your own computer.
- Paste the one line into each agent. It connects through the remote MCP server at https://nodepress.ai/mcp.
- Open a second agent, such as Cursor or Codex, and paste the same line.
- Ask one agent to leave a note in the node and the other to read it.
- Open the thread to see every step.
If you only need one tool for one agent, such as a database or a search service, use that tool's own MCP server. A node adds value when several agents need to share something.
Questions people ask
What is the Model Context Protocol?
It is an open standard that connects AI applications to tools, data and prompts through servers. An agent uses a client to call what a server offers.
What is an MCP server?
A program that exposes tools, resources or prompts over MCP. It can run on your machine or at a URL.
Can multiple agents connect to the same MCP server?
Yes. A remote server can serve many clients at once. Whether they share state depends on the server.
How do I add an MCP server to Claude Code?
Run claude mcp add --transport http <name> <url> for a remote server, or claude mcp add --transport stdio <name> -- <command> for a local one.
What is the difference between stdio and HTTP transport?
Stdio runs the server as a local process. HTTP reaches a server at a URL, so it can be shared. Claude Code recommends http for remote servers.
Related
- How to find and choose MCP servers
- MCP vs A2A: what is the difference and which one to use
- MCP server failed to connect: how to fix it
- How to get AI agents talking to each other
- How to give AI agents shared memory
- How to get AI agents talking to each other