What is a Remote MCP Server?

2 min read Updated

A remote MCP server is a hosted MCP server that clients reach over the network using the Streamable HTTP transport, rather than launching it as a local subprocess. Remote servers typically authenticate callers with OAuth-based authorisation and serve many users from one operated deployment.

WHY IT MATTERS

The MCP specification defines two standard transports: stdio, where the client launches the server as a subprocess, and Streamable HTTP, where the server runs as an independent process exposing a single MCP endpoint (e.g. https://example.com/mcp) handling HTTP POST and GET, optionally streaming responses via Server-Sent Events. Remote servers use the latter — the earlier HTTP+SSE transport from protocol version 2024-11-05 is deprecated and replaced by Streamable HTTP.

Authorisation differs sharply from local servers. The spec's authorisation framework is OAuth 2.1-based and applies to HTTP transports: clients obtain tokens via an authorisation server and present them on requests. Sessions are tracked with the MCP-Session-Id header. This gives operators real identity, token scoping and revocation — none of which exist for a subprocess inheriting its parent's environment.

The operational and security trade-offs cut both ways:

  • For remote: centrally patched and versioned, no code executing on user machines, credentials stay server-side, per-user authorisation and revocation, no supply-chain pull of npx packages at launch.
  • Against remote: a network dependency and shared attack surface; your data transits a third party's infrastructure; the operator can change tool behaviour at any time without anything changing locally — a variant of the MCP rug pull risk; and the spec requires servers to validate Origin headers and authenticate connections properly.

See remote mcp server working in your own stack — route your MCP servers through PolicyLayer and every tool call is checked against policy before it runs.

GOVERN YOUR MCP SERVERS →

Enforced before the call runs. Nothing to install.

HOW POLICYLAYER USES THIS

PolicyLayer's gateway fronts both remote and local-style upstreams: teams register a remote server's HTTP endpoint once, and clients connect through PolicyLayer with per-person scoped tokens instead of sharing upstream credentials. The public catalogue at /tools records per-tool risk classifications for thousands of MCP servers, remote and local alike.

FREQUENTLY ASKED QUESTIONS

What transport do remote MCP servers use?
Streamable HTTP — a single MCP endpoint supporting HTTP POST and GET, with optional SSE streaming. It replaced the deprecated HTTP+SSE transport from protocol version 2024-11-05.
How do remote MCP servers handle authentication?
Via the MCP authorisation specification, which is based on OAuth 2.1: clients obtain access tokens from an authorisation server and present them on requests to the MCP endpoint.
Are remote MCP servers safer than local ones?
They remove local code execution and credential exposure, but introduce a network dependency, third-party data handling, and the possibility of server-side behaviour changes. The risk profile is different, not strictly smaller.

FURTHER READING

Let agents act without letting them run wild.

Route your MCP servers through PolicyLayer and every tool call is checked against your policy before it runs — allow, deny, or require approval. Per-identity grants. Full audit log. Live in minutes.

Free to start. No card required.

43,000+ MCP servers and 220,000+ tools scanned and risk-classified.

// GET IN TOUCH

Have a question or want to learn more? Send us a message.

Message sent.

We'll get back to you soon.