Termcp

io.github.open-mcp-ai由 GitHub 账号 open-mcp-ai 发布v0.2.8更新于 Oct 27, 2026

An MCP server that lets AI agents act like a human at a real terminal, local or over SSH.

概览

AI 生成的概览

让 AI 代理像人一样操作真实终端的 MCP 服务器,支持本地或 SSH 远程,仅暴露一个工具。

可以做什么
  • 在本地或 SSH 远程的真实终端中执行命令
  • 在运行时安装、编写并热加载其他 MCP 服务器与技能
  • 管理主机,并通过 SSH 跨机器调用 MCP 服务器
  • 通过网页界面与人类共享同一批终端进行协作
适用场景
适合需要让助手直接操作 shell 而非调用单一 API 的场景:安装或接入其他 MCP 服务器、通过 SSH 在远程主机上执行命令,或与人类在共享终端会话中协同工作。
运行要求
以本地进程运行,通常通过 npx 加 daemon stdio 子命令启动,也可用 Go 安装、Docker 镜像或预编译二进制。已有实例在运行时可用 HTTP 传输。Docker 部署需设置 TERMCP_AUTH_TOKEN;使用 --auth-token 时客户端需发送 bearer Authorization 头。
安装前请注意
助手获得 shell 权限,可执行任意命令、修改或删除数据,并通过 SSH 操作远程主机。它还能安装并热加载其他 MCP 服务器,进一步扩大可执行范围。Docker 示例要求设置 TERMCP_AUTH_TOKEN,请使用足够长的随机密钥,并避免把 HTTP 端点暴露在不可信网络。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Termcp,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

[header banner]

[Termcp logo]

⚡ Termcp

[Termcp tagline]

[Stars] [Forks] [Release]

[Platform] [Pure Go, No CGO] [MIT License]

English | 中文

[Install] [Docs] [Website]

[divider]

Termcp: the final MCP | Terminus MCP

With Termcp:

  • An AI can install and use any number of MCP servers, dynamically, at run time.
  • An AI can write its own MCP server and hot-load it.
  • An AI can use SSH to manage hosts, approved and collaborated on by a human.
  • An AI can install and call MCP servers across machines over nothing but SSH.

Exactly one tool is exposed. Built on the Termcp protocol (Terminalization Model Context Protocol): loading more MCP servers and skills does not grow the context.

A single dependency-free binary (10 MB+), cross-platform, with a WYSIWYG web UI where a human and an AI share the same terminals and collaborate.

[Termcp extensions panel: which MCPs are installed, which host each one runs on, and whether it is available]

Install 80+ MCP servers and skills.

[Termcp console: local and SSH sessions on one screen, every line marked human / AI / shell output]

Host management.

Install

0 · AI install

Install npx -y @open-mcp-ai/termcp daemon stdio as an MCP named termcp.

Copy this to the AI.

1 · One command per agent

AgentCommand
Claude Codeclaude mcp add termcp -- npx -y @open-mcp-ai/termcp daemon stdio
Codexcodex mcp add termcp -- npx -y @open-mcp-ai/termcp daemon stdio
Gemini CLIgemini mcp add termcp npx -- -y @open-mcp-ai/termcp daemon stdio

2 · Config file

json
{  "mcpServers": {    "termcp": {      "command": "npx",      "args": ["-y", "@open-mcp-ai/termcp", "daemon", "stdio"]    }  }}
Other install methods — npm / Go / Docker / prebuilt binary
Command
npmnpm i -g @open-mcp-ai/termcp — installs a termcp command; with it, drop npx -y @open-mcp-ai/termcp from every snippet above and use termcp. Wrapper: open-mcp-ai/Termcp-npm.
Gogo install github.com/open-mcp-ai/termcp@latest — GOPROXY=https://goproxy.cn,direct in mainland China; also usable as a module (go get github.com/open-mcp-ai/termcp).
Dockerdocker run -d --name termcp -p 18765:18765 -v termcp-data:/home/termcp -e TERMCP_AUTH_TOKEN=change-me-to-a-long-random-secret ghcr.io/open-mcp-ai/termcp:latest termcp --no-internal --host 0.0.0.0 --port 18765
BinaryPrebuilt releases — linux / darwin / windows on amd64 / arm64, one static file each.

The npm wrapper downloads the matching prebuilt binary (override with TERMCP_VERSION, TERMCP_MIRROR, TERMCP_BIN, TERMCP_SKIP_DOWNLOAD). Flags: docs/cli.md. Docker recipes, Compose and a token-free loopback setup: docs/deploy.md.

HTTP transport, when an instance is already answering somewhere (npx -y @open-mcp-ai/termcp):

json
{  "mcpServers": {    "termcp": {      "type": "http",      "url": "http://localhost:18765/stream"    }  }}

Reachable from a client in Docker as http://host.docker.internal:18765/stream (macOS/Windows) or the host's LAN IP; from another container on the same network as http://termcp:18765/stream. Legacy SSE uses http://localhost:18765/sse — only /sse, JSON-RPC goes to /message by itself. With --auth-token, add a bearer header: "headers": {"Authorization": "Bearer <token>"}.

Docs

docs/cli.mdEvery flag and subcommand, capability gates, authentication.
docs/clients.mdAll four entrances end to end: MCP transports, the stdio bridge, authentication, REST examples, and the one-tool surface.
docs/deploy.mdDocker, multi-stage builds, Compose, pure-API build.
docs/protocol.mdTERM/1 specification: wire format, status codes, headers, the 19 directives, the three gates.
docs/capabilities.mdThe full capability surface and the security model.
docs/api.mdREST endpoints, WebSocket frames, locator resolution.
docs/capabilities.mdFull capability surface and the security model.
docs/architecture.mdSession kernel, entrances, storage.
docs/design/Design records: Web UI, resource model, session storage, font stack, mobile terminal.
CHANGELOG.mdRelease notes.

Benchmark

Comming Soon

Functional security testing

Comming Soon

Star History

[Star History Chart]

License

Released under the MIT License. Thanks to friends in the linux.do community for the discussions, feedback, and support.


[footer]

来源:README.md,提交 0595cb5

工具

0
SourceWeft 曾在隔离沙箱中以无凭据方式启动该软件包一次,以列出其工具,但未能完成:该软件包类型无法在沙箱中运行。

版本历史

9
  1. v0.2.8最新Oct 26, 2026
  2. v0.2.7Oct 9, 2026
  3. v0.2.6Oct 7, 2026
  4. v0.2.5Oct 1, 2026
  5. v0.2.4Sep 29, 2026
  6. v0.2.3Sep 26, 2026
  7. v0.2.2Sep 25, 2026
  8. v0.2.1Sep 20, 2026
  9. v0.1.17Sep 19, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分