Bible Network Crypto DeFi Onchain RWA AI Agent Stablecoin Chain SAFU CryptoTax DeFAI AGI Claude Me Claude Skill Claude Design Claude Cowork
Independent Media
Not affiliated with any project
Exploring the Frontier of AI Intelligence
claude-me.com
LATEST
OpenRouter Fusion API Launches: Three-Model Panel Nears Fable 5 Scores at Half the Cost — But Fable Itself Was Just Pulled by the US Government  ·  Getting Started with Claude Cowork: Hand a Whole Task to AI Without It Crashing at the Last Step  ·  Claude Code vs Cursor vs GitHub Copilot: Which AI Coding Tool Should You Actually Use?  ·  Turning Repeat Work into Reusable Skills in Claude: Stop Re-pasting the Same Long Instructions  ·  Build Your Own MCP Server: Safely Connect Claude to Your Internal Tools (With Permissions and Debugging)  ·  Claude Code Getting Started: Complete Flow from Installation to Your First Real Task
Glossary · mcp-tools

MCP Transport

mcp-tools 新手

30-Second Version · For the impatient
MCP Transport is the spec defining how Claude and an MCP server exchange messages. The two main options are stdio (standard input/output) and SSE/HTTP. stdio suits local connections on the same machine and is the simplest to configure; SSE/HTTP goes over the network and suits cloud deployments and team use. Which you pick depends on whether your server and Claude run on the same machine.
Full Explanation +
01 · What is this?

What do stdio and SSE actually mean?

stdio stands for standard input/output — the most basic way computer programs communicate. Picture an invisible pipe between two programs: one writes in, the other reads out. No network, no port, minimal configuration.

SSE stands for Server-Sent Events, a technique that keeps an HTTP connection open so the server can push messages to the client continuously. Think of it as keeping the line open so the server can speak whenever it has something to say — rather than the one-question-one-answer HTTP model. In MCP, SSE over HTTP lets Claude and a server communicate in real time across different machines, while credentials stay in your server and never reach Claude.

02 · Why does it exist?

After local development, do I have to switch from stdio to SSE/HTTP for team use?

Typically yes, but you don't have to rewrite logic from scratch. Most MCP server frameworks separate the transport layer from business logic, so switching transport is just swapping that layer — tool definitions and execution code stay untouched.

A few things to handle at the same time: add authentication (API key or OAuth), because any HTTP server is reachable by anyone who can route to it; enable HTTPS; confirm firewall or routing rules allow your intended users to connect. None of this is needed for local testing — none can be skipped for production. The final test: if someone without a valid token calls this server, does it reject them?

03 · How does it affect your decisions?

After setting up an SSE/HTTP server, how do I verify it is secure?

Security checks come in layers. First, authentication: verify that requests without a valid token are rejected outright, not just returned with an error. Second, authorization scope: passing auth doesn't mean access to everything — check that each token is limited to the tools and data range it should reach.

Third, input validation: don't trust that Claude's parameters are always clean; treat them like any external input. Fourth, logging: record every request — who called what tool, with what parameters, and what came back. With all four in place, you can trace the root cause of any problem.

04 · What should you do?

Advanced: is the MCP spec evolving and could transport change?

The MCP spec is led by Anthropic and still under active development. stdio and SSE/HTTP are the current main options, but the spec is extensible, so new transport layers can be added without breaking existing implementations.

The practical implication: code your transport choice in a way that's easy to swap rather than hard-coding it into business logic. MCP's core advantage is that tool definitions and transport are decoupled — conforming tools don't need rewriting when transport or version changes. Track announcements from Anthropic and the MCP community; breaking changes usually come with explicit migration guides.

Real-World Example +

Scenario: Chen just built an MCP server that lets Claude Desktop read a local SQLite database.

Early development: he uses stdio. Server and Claude Desktop run on the same Mac; three minutes to configure a path in claude_desktop_config.json, no network or auth setup. He tests this way for a month until the tool logic is solid.

When the team wants in: he switches the transport to SSE/HTTP, adds API key auth and HTTPS. Each team member configures the server URL and their own key in Claude Desktop. Not a single line of tool definition or query logic changes.

Diagram
stdio vs SSE/HTTP: two MCP transport options side by sideSide-by-side comparison of stdio (local, same-machine, pipe) and SSE/HTTP (network, cloud, multi-user) MCP transports with best-use notes.MCP Transport: stdio (local) vs SSE/HTTP (remote)stdio (local)Claude AppMCP Serverstdin/stdout pipeSame machineNo network or auth neededSimplest setup, one userBest for: local dev & personal toolsSSE / HTTP (remote)Claude AppMCP ServerHTTP requestSSE responseDifferent machines / cloudNeeds auth + HTTPSMulti-user, scalableBest for: team / cloud deploymentsClaude Me · claude-me.com
Feel free to share. Please credit the source.
Common Misconceptions +
✕ Misconception 1
x Myth 1: SSE/HTTP is always better because it sounds more advanced. Transport choice depends on use case. stdio is fastest to set up for local dev; SSE/HTTP's value is cross-machine and multi-user access. Using SSE/HTTP for a personal local tool just adds unnecessary complexity.
✕ Misconception 2
x Myth 2: Switching to SSE/HTTP automatically makes the server secure. Transport only determines how data moves, not who can call it or what they can do. An HTTP server on the internet is reachable by anyone who can route to it. Security requires authentication, authorization, and HTTPS — all separate from transport choice.
✕ Misconception 3
x Myth 3: Switching transport requires rewriting the whole server. A well-designed MCP server separates the transport layer from tool logic. Switching is just swapping the communication layer; tool definitions, queries, and return formats stay untouched.
The Missing Link +
Direct Impact

The core trade-off between stdio and SSE/HTTP is simplicity vs scalability.

stdio's advantage is zero friction: no network config, no auth, no HTTPS — live in three minutes. The cost is single-user, single-machine only.

SSE/HTTP's advantage is reach: the server can live anywhere and authorized users can connect. The cost is a full security layer you must take seriously — authentication, authorization, HTTPS, none optional. Choose by actual use case: personal tools get stdio; shared use gets SSE/HTTP. Don't add SSE/HTTP complexity just because it sounds more professional.

Ask a Question
Please enter at least 10 characters