How does an MCP server work?
The AI application, known as the MCP client, connects to the server and asks what it offers. The server answers with a list of tools, each with a name, a description and the inputs it accepts; it can also share resources (data to read) and prompts (reusable instructions). When a request calls for a tool, the model picks it and fills in the inputs, the client sends the call, and the server runs it and returns the result. Messages follow JSON-RPC, and a server runs on your own machine or remotely over HTTP. Before MCP, each AI application needed its own integration with each product; with one shared protocol, a single server works with every compatible client.
MCP server vs API: what's the difference?
An API is built for developers: someone reads its documentation and writes code against its endpoints. An MCP server is built for AI applications: it describes its own tools in a form a model can read, so a client can discover them and call them without integration code. Many MCP servers are a thin layer over an existing API, which still does the work. What the server adds is a choice: which actions an AI may take, in what shape, and under which checks.
Is it safe to let an AI act through an MCP server?
It depends on the server and on your client's settings. The MCP specification asks AI applications to get the user's consent before invoking a tool and to keep a person able to deny any call. Many clients also let you allow a tool once and for all, so the server's own limits matter: expose only the actions you mean to, scope each access key, put a confirmation step in front of anything hard to undo, such as publishing or deleting, and write every call to an audit log. That is human-in-the-loop design, applied to tools.
What does an MCP server look like in marketing automation?
In marketing automation, the tools map to the daily work: build a segment, draft a journey, read a report. fromHello's MCP server, for example, gives Claude, Claude Code, Cursor, Codex, Windsurf, VS Code or any MCP client 59 tools to create segments and templates, draft journeys and read analytics. No tool sends a message. A journey your AI builds stays a draft until it is published, and publishing, pausing or deleting through an agent takes a confirmation step in your AI client. Other edits, such as changes to a segment or a template, apply when saved, so they can reach a journey that is already live. MCP tool calls are recorded in your audit log.