MCP transports and Skill import
This is where ChengOS differs most from other workflow tools: it is both an MCP client for external tools and a place to package a capability into an installable Skill.
Three MCP transports
Stdio, SSE and HTTP are all supported. Stdio suits local process tools; SSE and HTTP suit remote services. They are parallel options in configuration — pick the one matching how the tool is deployed.
Five sources for a Skill
A Skill can be generated from a written requirement, or imported from web or API documentation, a Git repository, an existing SKILL.md, or a ZIP package. Import is a starting point, not a finish line: the result still needs human review, and arbitrary documentation sites are not guaranteed to parse cleanly.
How this meets credentials
Skills and MCP tools reach external services through credential handles, so secrets stay out of the model context. Revoking one credential cuts off every tool that referenced it.
Questions people ask
- Will an imported Skill just work?
- Not always. When importing from documentation or a web page, the result depends on how structured the source is, and needs review before you rely on it.
- Where do MCP tools run?
- It depends on the transport: Stdio tools run as local processes, while SSE and HTTP tools run on their own servers.
Evidence for the claims on this page
Last verified:
- 支持 MCP 的 Stdio、SSE 与 HTTP 三种传输方式。
- Scope: ChengOS 平台侧 MCP 客户端。
- Source: ChengOS MCP 传输实现 ·
- 支持从需求描述、网页/API 文档、Git 仓库、SKILL.md 与 ZIP 包导入 Skill。
- Scope: 导入结果仍需人工检查；不保证任意站点的文档均可解析。
- Source: ChengOS Skill 导入实现 ·
- 模型只能通过凭据句柄调用工具，API Key 明文不进入模型上下文。
- Scope: 凭据保管库与工具调用路径；不涵盖用户自行粘贴到对话中的内容。
- Source: ChengOS Credential Vault 与工具调用实现 ·