Remote vs local MCP servers

Every MCP server reaches your client in one of two ways: as a local process the client starts, or as a remote service it connects to over HTTP. The choice affects setup, what the server can reach, how you authenticate and what can go wrong.

The two standard transports

The MCP specification defines two standard transports. With stdio, the client launches the server as a subprocess and exchanges newline-delimited JSON-RPC messages over its standard input and output. With Streamable HTTP, the server runs independently and the client sends requests to a single HTTP endpoint. The server answers each request with a JSON object, or with a short-lived event stream when it needs to send several messages.

An older transport, HTTP with Server-Sent Events (SSE), has been replaced by Streamable HTTP. Some clients still support it for older servers, but new servers publish an /mcp-style Streamable HTTP endpoint. In this registry, listings marked HTTP use Streamable HTTP and listings marked stdio are local.

Local (stdio) servers

A local server is a program on your computer, usually started with npx, uvx or docker run. It runs with your user account's permissions, which is the point: it can read the folders you allow, drive a browser on your screen, or reach a database on your network that no cloud service can see.

  • Good for: files, local browsers, developer tools, databases inside a private network, and anything that must not leave your machine.
  • Setup: install the runtime the command needs, then put the command and any settings in your client's configuration.
  • Credentials: usually environment variables in the configuration, such as an API key or a connection string.
  • Risks: the code runs with your privileges. Install only from sources you trust, and pin or review what @latest installs if that matters to you.

Current local servers in the registry:

Remote (Streamable HTTP) servers

A remote server is a service someone else runs, usually the vendor of the product it connects to. You add a URL, and there is nothing to install or update. Remote servers are the natural fit for SaaS products, such as issue trackers, design tools, payment platforms and cloud consoles, because the data already lives there.

  • Good for: hosted products, sharing one setup across machines and clients, and anyone who doesn't want to manage runtimes.
  • Setup: add the URL to your client and complete sign-in.
  • Credentials: usually OAuth, so you approve access in a browser and the client stores a token. Some servers accept API keys in a header; a few public documentation servers need nothing at all.
  • Risks: your requests and the data the tools return pass through the provider's service. Read their terms, and check which account and permissions you grant during sign-in.

Current remote servers that use OAuth:

Current remote servers that work without authentication:

Current remote servers whose examples send a token or key in a header:

How to choose

  1. If the vendor runs an official remote server, start there. It is maintained by the people who run the API, it updates without you doing anything, and OAuth avoids long-lived keys in files.
  2. If the data is on your machine or inside your network, use a local server. A remote service cannot reach it, and you may not want it to.
  3. If both exist, compare what each offers. Some vendors put new features on the hosted server first; others keep advanced or self-hosted options in the local package.
  4. Prefer read-only modes where they exist. Several servers in the registry have them, such as a read-only flag or a read-only endpoint, and they are the easiest way to limit damage while you evaluate a server.

Local servers over HTTP

Some local packages can also listen on HTTP. The specification requires such servers to validate the Origin header to prevent DNS rebinding attacks, and recommends binding to 127.0.0.1 rather than all network interfaces. If you run a local server in HTTP mode, check that it does both, and add authentication if anything other than your own client can reach the port.

Set them up

The client guides show the exact configuration for each transport: Claude Code and Codex CLI, Cursor, VS Code and Claude Desktop. For tokens, permissions and approval prompts, read authentication and security basics.

Sources

This guide was written from the official documentation below and last reviewed on 12 October 2026. Client interfaces change often; when a menu or file name here differs from what you see, the upstream documentation wins. Report a correction through the contact page.