MCP servers talk to AI clients over a transport. Two matter in practice: stdio for local servers and Streamable HTTP for remote ones. The choice changes how you deploy, secure and maintain the server.
Local servers (stdio)
The AI client starts the server as a process on the user's computer and talks to it through standard input and output.
- Good for: pilots, developer tools, anything that needs local files.
- Downsides: every person installs and updates it, credentials live on every laptop, and there is no central log.
Remote servers (Streamable HTTP)
The server runs in the cloud or your network at one URL. Clients connect over HTTPS, and users usually sign in with OAuth.
- Good for: teams, company data, anything that needs permissions and audit.
- Downsides: you need hosting, TLS, authentication and monitoring.
Comparison
| Local (stdio) | Remote (Streamable HTTP) | |
|---|---|---|
| Setup per user | Install and configure | Add a URL and sign in |
| Credentials | On each machine | On the server only |
| Per-user permissions | Hard | Natural with OAuth |
| Audit log | Scattered | Central |
| Updates | Everyone updates | Deploy once |
Our rule of thumb
Pilot locally with two or three people. Once the tools are right, move to a remote server before rolling out to the team. Building both from the same code base is easy if you plan for it from the start.
See how we build and host remote servers: MCP hosting and maintenance.