适配器上下文和策略辅助工具

此 API 为 DCC 适配器提供共享契约，用于简洁指令、工具后快照、可视反馈、响应整形、工具集配置文件和可搜索的 API 文档。

指令资源

from dcc_mcp_core import AdapterInstructionSet, register_adapter_instruction_resources

register_adapter_instruction_resources(
    server,
    AdapterInstructionSet(
        dcc="maya",
        instructions="Use screenshots after visual scene changes.",
        capabilities={"screenshots": True, "in_process_execution": True},
        troubleshooting="If execution stalls, check for modal dialogs.",
        adapter_version="0.3.0",
    ),
)

这将注册：

• docs://adapter/<dcc>/instructions
• docs://adapter/<dcc>/capabilities
• docs://adapter/<dcc>/troubleshooting（提供时）

DccServerBase.register_adapter_instructions(...) 是嵌入式适配器的薄包装器。

上下文快照

适配器可以在变异工具后附加一个小的、有界的场景/文档快照：

from dcc_mcp_core import DccContextSnapshot, append_context_snapshot

result = append_context_snapshot(
    {"success": True, "message": "Layer created"},
    DccContextSnapshot(
        dcc="photoshop",
        document={"name": "hero.psd"},
        active_layer={"name": "Glow"},
        counts={"layers": 12},
    ),
)

DccServerBase.set_context_snapshot_provider(callable) 存储提供者，DccServerBase.append_context_snapshot(result) 应用它。

响应整形

ResponseShapePolicy 截断大型列表、字典和字符串，并添加带有省略计数和 next_query 提示的 _meta["dcc.response_shape"]。

from dcc_mcp_core import ResponseShapePolicy, shape_response

payload = shape_response(scene_graph, ResponseShapePolicy(max_items=200, max_bytes=256_000))

可视反馈

VisualFeedbackPolicy 标准化资源支持的预览：

from dcc_mcp_core import VisualFeedbackPolicy, build_visual_feedback_context

context = build_visual_feedback_context(
    resource="output://viewport.png",
    width=1280,
    height=720,
    policy=VisualFeedbackPolicy(mode="after_mutation", max_size=800),
)

工具集配置文件

DccToolsetProfile 和 ToolsetProfileRegistry 在技能组之上提供高级层，用于适配器模式，如 modeling-basic、rendering 或 photoshop-layer-editing。