What Is a Remote MCP Server and How to Connect It
A remote MCP server runs your tools in the cloud so agents connect over HTTP with no local setup. Learn how remote MCP differs from local, why it scales better, and how to connect it to Claude Code, Cursor and VSCode.

Tool names in this guide are for reference only and may have changed or been renamed. Please use the latest tools in the tool marketplace when implementing.
Key takeaways
- Understand the difference between a local MCP server and a remote one
- Know why remote MCP removes install, update and scaling overhead
- Connect a remote MCP server to Claude Code, Cursor or VSCode in one step
Tools used
What is a remote MCP server?
An MCP server exposes a set of tools that an AI agent can call through the Model Context Protocol. Where that server runs is what splits the world into two kinds.
A local MCP server is a process you launch on your own machine — usually over stdio — so the agent talks to it through standard input and output. A remote MCP server runs on infrastructure you don't manage and the agent reaches it over the network, typically via Streamable HTTP with an API key. Same protocol, different location: one lives next to your editor, the other lives in the cloud.
That single difference changes almost everything about setup, maintenance and scale.
Local vs remote MCP, side by side
| Dimension | Local MCP server | Remote MCP server |
|---|---|---|
| Where it runs | Your machine (stdio) |
Cloud, over HTTP |
| Install & runtime | You install and launch it | Nothing to install |
| Updates | You pull new versions | Updated server-side |
| Credentials | Per-tool keys on your disk | One key to the provider |
| Tool count | What you host locally | Whatever the provider offers |
| Sharing across machines | Reconfigure each device | Same endpoint everywhere |
Local servers are great for a single tool you own or for offline development. Remote servers win the moment you want many tools, consistent uptime, and the same setup on every laptop and CI runner.
Why teams reach for remote MCP
No ops. You don't install packages, pin runtimes, or babysit a process. The endpoint is just there, and upgrades land server-side without touching your config.
Many tools behind one door. A remote server can front hundreds of capabilities — financial data, web search and scraping, image and video generation, maps, e-commerce and social data — all reachable through one connection and one key instead of a dozen local integrations.
Portable setup. Because the agent only needs a URL and a key, the exact same configuration works on your laptop, a teammate's machine, and an automated agent in CI.
How to connect a remote MCP server
The flow is the same across clients: point the agent at the server's HTTP endpoint and give it your key.
- Get an endpoint and key. With AgentEarth you sign up and new users get 1,000 free credits; you receive one API key that unlocks every tool.
- Connect in-app or by config. In agents like WorkBuddy or Trae, search for "AgentEarth" and connect in-app. For Claude Code, Cursor or VSCode, paste the one-line remote MCP config with your key.
- Just ask in natural language. Say use ae to pull the latest quote for AAPL or use ae to scrape example.com and summarize it — the agent discovers the right tool from the remote server and runs it. No per-tool wiring.
Because discovery is part of the protocol, you never hard-code tool names: the agent lists what the remote server offers and picks the match for your request.
Once connected, the tools run remotely: ask the agent to search the web and it calls the search tool on the server (such as ae_serper_search) — nothing to install on your machine.
Common pitfalls
- Transport mismatch. A remote server speaks HTTP, not
stdio. If your client config is still set to launch a local command, it will never reach the endpoint — use the remote/URL style config. - Missing or wrong key. Remote servers authenticate every call. A 401 almost always means the API key is absent, mistyped, or not passed in the header the client expects.
- Network and firewall. Corporate proxies or egress rules can block the endpoint. If connections are refused, confirm the host can reach the server URL before blaming the config.
FAQ
Is a remote MCP server better than a local one? For most people, yes — it removes install and maintenance work and gives you far more tools through one connection. Local servers still make sense for a single self-hosted tool or offline work.
Is remote MCP secure? Access is gated by an API key over an encrypted connection, and you send only the parameters a call needs. Treat your key like any other secret and rotate it if it leaks.
Which clients support remote MCP? Claude Code, Cursor, Codex, Trae, VSCode and WorkBuddy, plus any MCP-compatible agent that supports HTTP transport.
Get started
- Sign up — new users get 1,000 free credits.
- Connect in-app (WorkBuddy / Trae) or paste the remote MCP config into Claude Code / Cursor / VSCode.
- Give instructions in plain language and let the agent do the rest.
