SuperAPI AI Directory

dcc-mcp-creator

Agents & models

View repository

Updated 10-16 08:35

About this skill

DCC-MCP Creator

Implement the requested adapter or service within its owner's project. A private service needs no public repository, catalog entry, issue or release. For operating an existing app use dcc-mcp; for individual tool packages use dcc-mcp-skills-creator. An already-loaded skill needs no installation or new agent turn.

Implementation boundaries

Use DccServerBase and DccServerOptions.from_env(...), with host API calls through HostExecutionBridge. Embedded, sidecar and standalone services have different owner/host lifetimes; resolve that contract before wiring the runtime. Reuse Core's public lifecycle, discovery and state owners rather than parallel adapter implementations. Keep identity and host configuration data-driven.

Application UI uses project-owned dcc-cua / ui-control. Before observation or input report provider=dcc-cua runtime=<version> pid=<exact-pid> hwnd=<exact-native-hwnd>. Missing binding blocks actions. Read the bundled dcc-cua skill for UI work; never fall back to generic Computer Use without the user's explicit change of provider.

Choose relevant guidance

Work Read
New or modernized public adapter Adapter workflow, host patterns
Private standalone service Internal service workflow
Server ownership, host dispatch, state digest, UI integration, gateway or sidecar wiring Runtime integration; consult its relevant numbered contracts
Proposed adapter-local workaround for shared behavior Core escalation
Async/main-thread jobs, cancellation or reconnect Async recovery
Runtime failure or issue ownership Failure routing
Nuke/standalone examples or cross-DCC file synchronization Integration examples
Validation or release Testing and release

Use the dcc-mcp route only when live validation is needed; source-only edits do not require starting a host, installing a CLI, or updating a server. Run the relevant local checks, fix failures caused by the change, and verify the requested behavior before reporting completion. Keep source/mock validation separate from real-host acceptance. Publishing requires the requested scope and the applicable release gates; native Python 3.7 remains an LTS profile, and py37-lite does not satisfy its release gate.

Non-Negotiables

  • Do not touch a DCC API from a Tokio/HTTP worker thread.
  • Do not parse or rewrite SKILL.md, tools.yaml, groups.yaml, or prompt/workflow files in adapter runtime code when core exposes a typed object or catalog API.
  • Do not reach into server._server unless no public core API exists; if you must, file a core issue and keep the adapter shim small.
  • Do not create Maya-only abstractions in shared core or adapter templates.
  • Do not expose raw script execution as the primary user workflow when a typed skill can cover the task.
  • Do not require GitHub, a public catalog entry, or public issue tracking for a private internal service.
  • Do not publish local paths, private machine names, or source-attribution markers in public issues or PR text.