glossary · Running it

MCP client

definition

An MCP client is the part of an AI application that connects to MCP servers, discovers their tools and resources, and passes the model's calls to them, so the model can act on systems outside the chat.

in one line: The part of your AI tool that talks to MCP servers.

in context

Where does an MCP client fit?

Keep your client. Add the process. In ConvOps, Claude Code, Codex, Cursor, ChatGPT and OpenCode reach the same server and the same steps.

Claude Codeand Claude, Cowork
Codexand ChatGPT
Cursorand OpenCode
Clients verified with ConvOps. Any MCP client can connect.
two ways to say it

What is an MCP client, in plain words?

Same idea at two depths: the plain version, then what the engine actually does.

Your AI tool is the host. Inside it, a client holds each server connection and passes the model's calls along.

Switch tools and keep the same servers.

for engineers

claude mcp add --transport http convops https://mcp.convops.app/

Most other clients take the same URL in their MCP settings and open a browser to sign in.

the longer answer

Why does an MCP client matter?

In the Model Context Protocol, the application a person uses, such as an IDE, a chat app or a terminal agent, is called the host. Inside it, each connection to a server is held by a client. The client asks the server what it offers, shows those tools to the model, sends the model's tool calls to the server, and hands the results back. The model never talks to the server directly.

That separation is what makes MCP portable. A team can switch clients, or use several at once, and keep the same servers. A developer in a terminal agent and a colleague in a desktop chat app can both reach the same system with the same tools, under their own sign-in.

Clients differ in what they support. Some connect only to local servers started as processes, others also to remote servers over HTTP. Some handle OAuth sign-in in a browser, others need a token. Some expose resources and prompts as well as tools. When a server does not work in a client, the cause is usually one of those three: transport, authentication, or which capabilities the client implements.

in convops

How does an MCP client work in ConvOps?

ConvOps is reached through an MCP client; that client is where your work happens. Clients verified to work with ConvOps are Claude Code, Claude on desktop and web including Cowork, OpenAI Codex, ChatGPT, Cursor and OpenCode, and the server follows the protocol so other MCP clients can connect too. In Claude Code the install is `claude mcp add --transport http convops https://mcp.convops.app/`; most other clients take the same URL in their MCP settings. Your client keeps its model and subscription. ConvOps supplies the process: the current step, its gates, the workspace memory and the record. That is attended execution: you drive in your client, and the engine supplies the steps.

questions

What do people ask about an MCP client?

What is an MCP client, and how does ConvOps use one?

An MCP client is the part of an AI application that connects to MCP servers, discovers their tools and passes the model's calls to them. ConvOps is reached through one: the client is where your work happens, and ConvOps supplies the current step, its gates, the workspace memory and the record. Your client keeps its own model and subscription.

What is the difference between an MCP client and an MCP host?

The host is the application you use, such as an IDE, a chat app or a terminal agent, and the client inside it holds each connection to one server. With ConvOps, Claude Code is the host and its MCP client holds the ConvOps connection, discovering the tools and relaying the model's calls. The model never talks to the server directly.

Which MCP clients work with ConvOps?

Verified clients for ConvOps are Claude Code, Claude on desktop and web including Cowork, OpenAI Codex, ChatGPT, Cursor and OpenCode. Other clients that follow the protocol can connect too, and most take the same URL, https://mcp.convops.app/, in their MCP settings. A team can mix clients and still work one shared process.

Why does an MCP server not work in my client?

Usually transport, authentication or capabilities: the client may support only local servers, may need a token instead of OAuth sign-in, or may not implement resources and prompts. ConvOps is a remote server with OAuth sign-in, and API keys for clients without it, so check first that your client supports remote HTTP servers.

Which people already have an MCP client without knowing it?

Most people who use Claude Code, Claude, ChatGPT, Cursor, Codex or OpenCode already have one, built into the app. ConvOps is reached through that client, so anyone on a team can join the shared process from the tool they prefer, without installing a separate application or changing model provider.