If you are deciding whether this is meaningfully different from a normal SSH wrapper, this page gives the clearest side-by-side explanation.很多用户在信任一个新的远程运行时之前，会先做对比搜索。这个页面就是专门回答：当终端变成“共享的、有状态的、对人和 AI 同时可见的”之后，到底改变了什么？
The strongest argument is behavioral, not marketing. The comparison below stays concrete.最有说服力的不是营销话术，而是行为差异。下面的对比尽量保持具体。
Each command is treated like a separate call. Prompt state, partial context, interactive pauses, and manual takeover become awkward because the wrapper is not centered on a durable PTY.每条命令都像单独调用。提示符状态、半截上下文、交互暂停和人工接管都会变得尴尬，因为它不是围绕一条持久 PTY 来设计的。
The browser viewer, input lock, async tracking, diagnostics, and policy rules all make more sense once the terminal is treated as a shared, stateful place instead of disposable command output.一旦终端被当成共享的、有状态的地方，而不是一次性命令输出，浏览器 viewer、输入锁、异步跟踪、诊断和策略规则就都变得合理了。
Each card focuses on one practical difference you can feel in a real workflow.每张卡都只讲一个你在真实工作流里能感受到的差异。
Shared PTY keeps shell state visible to both sides instead of reconstructing context after every command.共享 PTY 会把 shell 状态留给双方，而不是每次命令后再去重构上下文。
A viewer gives the human a live window into prompts, logs, and terminal anomalies.viewer 会把提示符、日志和终端异常实时暴露给人。
Input lock matters when the shell turns interactive or risky and the AI must stop typing over the operator.当 shell 变成交互态或高风险态时，输入锁能阻止 AI 覆盖操作者输入。
Async tracking acknowledges that remote work is often slow and non-blocking.异步跟踪承认远程工作本来就经常是慢的、非阻塞的。
Policy rules let operators expose AI help without fully surrendering the machine.策略规则让操作者能开放 AI 帮助，但不必把整台机器完全让出去。
If you are still judging whether this is meaningfully different, start with these.如果你还在判断它是不是和普通做法有本质区别，可以先看这几条。
The biggest difference is that SSH Session MCP keeps one shared PTY alive for the human and AI, while a normal wrapper usually treats each command like isolated, disposable work.最大的差异是：SSH Session MCP 会为人和 AI 保留一条共享 PTY，而普通 wrapper 往往把每条命令都当成相互隔离、可丢弃的单次执行。
Because visibility changes behavior. A visible terminal is easier to trust, inspect, interrupt, and manually take over than hidden command execution.因为“可见性”本身就会改变行为。一个可见终端比隐藏式命令执行更容易被信任、检查、中断和人工接管。
Once the terminal becomes shared and stateful, you need a way to stop the AI and human from racing each other in the same shell. Input lock is that control.一旦终端变成共享且有状态，就必须有办法阻止 AI 和人同时在同一个 shell 里互相抢输入，而输入锁就是这个控制。
Because a stateful remote runtime acknowledges that commands may outlive one request-response turn. Async tracking makes that explicit instead of pretending everything finishes immediately.因为有状态的远程运行时承认：命令可能会活得比一次请求响应更久。异步跟踪把这件事显式表达出来，而不是假装所有命令都会立刻结束。
From here, the next step is usually install steps, examples, or a few direct answers.从这里继续，下一步通常就是去看安装、例子或几个直接答案。