ETF Agent adds a Tool Surface for verifying model-visible tools
ETF Agent developers consolidated business-tool definitions under MCP and introduced Model Capability Presentation and Tool Surface to verify the final logical tool list sent to a model. The new layer derives its expected tools from existing authorization facts and checks them after hooks, context processing, evidence assembly, and asynchronous preparation, while rejected dispatches are settled through the existing request owner. Controlled regression tests covered 13 counterexamples, according to the developers, but the report does not independently validate production behavior.
The approach addresses a concrete agent-engineering problem: a tool may be registered, authorized, or declared by a skill without being legally visible or usable in the final model request.
给 Agent 接入 MCP 工具，通常要做几件事：注册工具、定义 Schema、配置权限，再把工具交给模型调用。听起来并不复杂。
但开发 ETF Agent 时，我们先后遇到过两种相反的问题：系统已经有某个工具，模型却看不到；模型能看到某个工具，Host 却未必真正允许执行。
为了解决这些问题，我们先统一工具定义，再统一工具暴露规则，后来又把 Runtime 里重复的业务工具定义删掉，让 MCP 成为真正的能力接口。
这些重构都解决过实际问题。可是到了 2026 年 10 月，我们才意识到：即使前面的授权、工具注册和能力投影都正确，也不能保证经过 Hook、上下文处理和异步准备之后，模型最终收到的工具列表仍然正确。
于是有了模型能力呈现与工具表面（Model Capability Presentation & Tool Surface）。
它不是一次从零开始的架构设计，而是几轮重构后留下的最后一块责任空白。