Client setup

MCP, set up
for your client.

A remote MCP server is one URL. The setup is different in every client. Pick the software you use for the exact command, config path, and approval step.

14client guides
3,808remote servers
1setup pattern
Start here

The most-used MCP clients

These clients support remote MCP servers directly. Each guide shows the project and global config paths, the setup shape, and the approval behavior.

All guides

Choose your client

Every guide is grounded in the client's current documentation. Native setup steps are shown where a client has its own CLI; the rest use the client's documented config shape. If a client is global-only, the guide says so instead of inventing a project path.

The fast path

Install a server from its page

Every server page has a client selector. It shows the config shape for the real endpoint, server name, and transport. Use this shape when you already have a URL.

OpenCode V2 · project
opencode mcp add example --url https://mcp.example.com/mcp
Claude Desktop · JSON
{
  "mcpServers": {
    "example": {
      "url": "https://mcp.example.com/mcp"
    }
  }
}
  1. 1Find the server in the MCP directory.
  2. 2Choose your client in the sidebar and copy the setup details.
  3. 3Restart the client, authorize if asked, and ask for a task.
How MCP works

One protocol, different client plumbing

MCP is the interface between a client and a server. Each client still has its own config file, scope rules, and approval model.

01

The server advertises tools

A remote MCP server exposes tools, resources, and prompts. The client discovers them when it connects; you do not write tool definitions by hand.

02

The client holds the config

The URL lives in the client's own config file. Some clients support project and global scopes; others are global-only or read a single shared file.

03

The provider controls access

OAuth, API keys, rate limits, and tool permissions belong to the provider. The setup guide links to the provider's docs rather than guessing.