DCC-MCP Agent Plugins

The official distribution repository for DCC-MCP Agent Skills across Codex, Claude Code, CodeBuddy, WorkBuddy, OpenClaw, ClawHub, Gemini CLI, GitHub Copilot, Cursor, Windsurf, OpenCode, Cline, Roo Code, Kiro, Amp, and other Agent Skills-compatible hosts.

One canonical Skill suite lives under plugins/dcc-mcp/skills/. Vendor manifests are thin adapters; they do not fork instructions or runtime behavior. Released-product discovery is generated from plugins/dcc-mcp/skills/dcc-mcp/references/PRODUCTS.json, a provenance-bearing snapshot of the official dcc-mcp-cli dcc-types catalog reconciled with the owning adapter repositories. It also carries a small current-Core route set for adapters that have landed in Core before the next CLI release. It carries one canonical identity and bounded aliases per route instead of repeating hand-maintained product lists across manifests.

Install

Most agents (recommended)

Install the default dcc-mcp Skill with one command. The installer detects supported agents in the current workspace and lets you choose the target:

npx --yes skills@1.5.22 add dcc-mcp/dcc-mcp-agent-plugins --skill dcc-mcp

This covers Gemini CLI, GitHub Copilot, Cursor, Windsurf, OpenCode, Cline, Roo Code, Kiro CLI, Amp, OpenHands, Continue, Replit, and other Agent Skills-compatible hosts. Start a new agent session if the host discovers Skills only at startup. Use --global for a user-level installation or --agent <id> --yes for a non-interactive explicit target; see the installer's supported-agent list.

After installation, prompts only need to describe the work:

Use the dcc-mcp Skill to <describe the DCC task>.

The Skill handles DCC discovery, typed tool selection, and approval boundaries. Install dcc-mcp-creator or dcc-mcp-skills-creator only when developing an adapter or authoring a reusable DCC-MCP Skill.

Portable Agent Plugin

GitHub Release archives include a root plugin.json conforming to the Agent Plugins 1.0.0 specification and the same canonical skills/ directory. Compatible clients can load the archive as a skills-only portable plugin; installation and distribution remain client-defined by the standard.

Native plugin marketplaces (optional)

Use a native marketplace when the host provides one. It installs the same canonical Skills.

Codex

codex plugin marketplace add dcc-mcp/dcc-mcp-agent-plugins
codex plugin add dcc-mcp@dcc-mcp

Claude Code

/plugin marketplace add dcc-mcp/dcc-mcp-agent-plugins
/plugin install dcc-mcp@dcc-mcp

CodeBuddy

codebuddy plugin marketplace add dcc-mcp/dcc-mcp-agent-plugins
codebuddy plugin install dcc-mcp@dcc-mcp

WorkBuddy

Add this repository URL from Plugins > +, or upload a standalone Skill ZIP from the GitHub Release to Skills > Add skill > Upload skill.

OpenClaw and ClawHub

openclaw skills install @loonghao/dcc-mcp
openclaw skills install @loonghao/dcc-mcp-skills-creator
openclaw skills install @loonghao/dcc-mcp-creator

Direct ClawHub installs are also supported:

npx --yes clawhub@0.23.1 install @loonghao/dcc-mcp
npx --yes clawhub@0.23.1 install @loonghao/dcc-mcp-skills-creator
npx --yes clawhub@0.23.1 install @loonghao/dcc-mcp-creator

Keep installed Skills current

Use the updater that owns the installation. OpenClaw records ClawHub installs, so workspace and shared Skills can be refreshed non-interactively:

openclaw skills update --all
openclaw skills update --all --global

A direct ClawHub install is tracked in the work directory where it was installed. Run the update from that same directory:

npx --yes clawhub@0.23.3 update --all

Skills installed from GitHub with the universal Agent Skills CLI use its own lock file and updater:

npx --yes skills@1.5.23 update -p -y
npx --yes skills@1.5.23 update -g -y

These non-interactive commands can run from Task Scheduler, cron, or the host's automation service. Keep --force out of unattended ClawHub updates so pinned or locally modified Skills stay protected. The universal Agent Skills updater reinstalls changed managed copies from their recorded source, so custom changes belong in the source repository. An untracked directory copied into an agent's Skill folder must be reinstalled once through one of these managers before automatic updates can discover it.

Smithery Skills

The three canonical Skills are mapped to Smithery's GitHub-backed Skills Registry. Search and install them with the Smithery CLI:

npm install -g smithery@latest
smithery skill search "dcc-mcp"
smithery skill add loonghao/dcc-mcp --agent claude-code

Maintainers can validate the mapping locally, then publish only with an explicit flag and SMITHERY_API_KEY:

python scripts/smithery_sync.py
SMITHERY_API_KEY=... python scripts/smithery_sync.py --publish

npm

The same canonical Skills ship as a package for hosts that vendor Skills through node_modules:

npm install @dcc-mcp/skills
npx --yes skills@1.5.22 experimental_sync

PulseMCP

PulseMCP is an MCP Server directory, not a raw Agent Skills registry. This repository therefore does not publish Skill ZIPs there. Submit a hosted or local DCC-MCP Server from the owning adapter/core release through pulsemcp.com/submit; keep Agent Skills on Smithery, ClawHub, or an Agent Skills-compatible host.

Catalog and discovery

Every release publishes a machine-readable catalog and GEO metadata to GitHub Pages, generated from the same manifests CI validates:

File	Purpose
catalog.json	Skill metadata, install commands, and distribution channels
llms.txt	Curated entry point for AI crawlers and answer engines
llms-full.txt	Every SKILL.md concatenated in full
sitemap.xml	Per-Skill pages carrying schema.org/SoftwareApplication JSON-LD

catalog.json also projects the complete released-product matrix and the application UI route. Typed DCC-MCP tools remain first. DCC-CUA and ui-control are searchable names for the same project-owned application UI provider across DCC, browser, and non-DCC apps; explicit DCC-CUA requests never fall back to generic Computer Use, Browser, or Chrome providers.

docs/DISTRIBUTION.md is generated from the same source and lists every channel plus ready-to-paste entries for the directories that need a human submission. CI fails when it drifts:

python scripts/build_geo_site.py          # regenerate site/ and docs/DISTRIBUTION.md
python scripts/build_geo_site.py --check  # fail when the committed doc is stale

Skill suite

Skill	Purpose
dcc-mcp	Operate live DCC applications through structured tools
dcc-mcp-skills-creator	Create and validate DCC-MCP Skills
dcc-mcp-creator	Create and modernize DCC-MCP adapters

Skill instructions are authored in dcc-mcp-core, next to the runtime they document; this repository owns distribution — product discovery metadata, vendor manifests, archives, GitHub Releases, ClawHub, and Smithery. Content is copied by scripts/sync_core_skills.py from the Core commit pinned in .github/core-skills-sync.json, and CI fails when the two sides drift. See docs/adr/0002-skill-sync-direction.md.

The sync preserves only bounded distribution-owned data: each Skill version, plus the default Skill's generated description, search hint, tags, OpenAI interface, marked product/UI routing block, and PRODUCTS.json. Core continues to own the remainder of the instructional body and every other file. Conditional DCC-CUA routing is deliberately not declared as a hard depends edge, so typed-tool tasks keep progressive loading and do not load the UI provider unless UI behavior is actually required.

The Skill suite version is this repository's own release line, deliberately decoupled from dcc-mcp-core release numbers: Skill content can be re-published without a Core release, and Core can release without changing Skill content. metadata.dcc-mcp.version in each SKILL.md is therefore the one field the sync preserves instead of copying.

Core skill synchronization and product discovery are separate contracts. The daily core-sync.yml job compares the three canonical Skill bodies with the pinned Core checkout. Product discovery is regenerated from PRODUCTS.json and can add a route from the current Core catalog (for example OBS Studio, LiquiGen, or the Office/PPT workflow) even while the released CLI snapshot remains at its own version. Run sync_product_discovery.py --check --check-core-catalog to verify both the generated surfaces and the pinned current catalog.

When a user gives an absolute local application path, the default Skill stores it in the per-user cache used by scripts/app_path_cache.py. Later requests reuse the path only after checking that it still exists and asking the user to confirm startup. Missing paths produce a new-path prompt and an install hint; no process is started automatically. See LOCAL_APP_PATH_CACHE.md.

# Re-sync after Core changes Skill content, then bump the suite version.
python scripts/sync_core_skills.py --source ../dcc-mcp-core
python scripts/bump_version.py --patch
python scripts/sync_product_discovery.py
python scripts/build_geo_site.py
python scripts/sync_core_skills.py --source ../dcc-mcp-core --check
python scripts/sync_product_discovery.py --check --check-core-catalog --check-cli

core-sync.yml runs that sequence itself every day and on demand. When Core has published newer Skill content it re-syncs, bumps the suite version, regenerates the distribution metadata, and opens or updates a pull request on automation/core-skill-sync. A Core commit that changed no Skill content opens nothing. Merging any validated Skill version change publishes the three Skills to ClawHub. Tagging v<version> still creates the cross-channel GitHub Release, Smithery, npm, and Pages release.

Set an AUTOMATION_TOKEN or organization PERSONAL_ACCESS_TOKEN secret (a PAT with contents and pull-requests write access) so the generated pull request also triggers Validate; the default GITHUB_TOKEN cannot start workflows from its own commits. Without it the pull request is still opened, just without CI runs.

Development

python scripts/setup_released_core.py
python -m unittest discover -s tests -v
python scripts/sync_product_discovery.py --check
python scripts/validate_repository.py
npx --yes skills@1.5.22 add . --list
./scripts/build-packages.ps1

codex plugin marketplace add .
codex plugin add dcc-mcp@dcc-mcp

claude plugin validate . --strict
codebuddy plugin validate ./plugins/dcc-mcp

Tagging v<version> creates a GitHub Release with the plugin and standalone Skill archives, then fans out to every automated channel: ClawHub, Smithery, npm, and the GitHub Pages catalog. A final matrix job installs the released Skills from the public repository on Linux, macOS, and Windows and asserts the released version, so a broken public install fails the release.

Secret	Channel
CLAWHUB_TOKEN	ClawHub
SMITHERY_API_KEY	Smithery
NPM_TOKEN	npm (@dcc-mcp/skills, published with provenance)

GitHub Pages must be set to Build and deployment > Source: GitHub Actions. skills.sh has no publish API — it indexes public repositories and ranks them from anonymous skills CLI telemetry — so CI verifies public installability and never generates install telemetry.

OpenAI and Claude public directories require human review; CI builds and validates their submission artifacts but does not bypass those review portals. See SUBMISSION.md.

Security boundary

The plugin ships no public MCP server. It connects to the user's local DCC-MCP gateway through dcc-mcp-cli; a public multi-tenant MCP endpoint is a separate security and deployment product.

Product catalog, package, and CI evidence proves discoverability and artifact parity only. It does not claim licensed real-host validation for any listed application.

License

Repository code is MIT. Published Skill content is MIT-0 as declared by each SKILL.md and required by ClawHub.