From ark-ai-mcp
Query public GitHub repos via DeepWiki MCP — AI-powered answers about architecture, internals, and design decisions without cloning. Use for: 'how does this repo work', 'explain the architecture of X', 'compare two frameworks'. No auth required.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ark-ai-mcp:mcp-deepwikiThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
<objective>
DeepWiki vs Context7 vs GitHub MCP:
For understanding an unfamiliar repo, follow this order:
See what documentation topics are available:
mcporter call 'deepwiki.read_wiki_structure(repoName: "facebook/react")' --output json
Returns a table of contents for the repo's wiki. Use this to understand what's documented before asking questions.
Best for specific questions. Returns AI-powered, context-grounded answers:
mcporter call 'deepwiki.ask_question(repoName: "facebook/react", question: "How does the fiber reconciler work?")' --output json
Gets the complete documentation for a repo. Warning: Can return 10K+ tokens. Only use when you need a comprehensive dump:
mcporter call 'deepwiki.read_wiki_contents(repoName: "facebook/react")' --output json
Prefer ask_question for targeted queries — it's faster and returns only relevant content.
Compare approaches across repos (max 10):
mcporter call 'deepwiki.ask_question(repoName: ["vercel/next.js", "remix-run/remix"], question: "How do these frameworks handle server-side rendering differently?")' --output json
# Understand a library's architecture
mcporter call 'deepwiki.ask_question(repoName: "steipete/mcporter", question: "How does the config discovery and merging work?")' --output json
# Investigate design decisions
mcporter call 'deepwiki.ask_question(repoName: "anthropics/claude-code", question: "How does the permission system work?")' --output json
# Compare competing libraries
mcporter call 'deepwiki.ask_question(repoName: ["expressjs/express", "fastify/fastify"], question: "How do these handle middleware differently?")' --output json
# Explore what docs exist before deep diving
mcporter call 'deepwiki.read_wiki_structure(repoName: "langchain-ai/langchain")' --output json
- **Start with `read_wiki_structure`** to see what's available, then use `ask_question` for specifics.
- **`ask_question` is your primary tool** — it returns focused, relevant answers without the token cost of a full wiki dump.
- **`read_wiki_contents` is expensive** — it returns the entire wiki (often 10K+ tokens). Only use it when you need comprehensive coverage.
- The `repoName` format is always `owner/repo` (e.g., "facebook/react").
- For multiple repos, pass an array (max 10): `repoName: ["repo1", "repo2"]`
- No authentication required — works out of the box.
- Best for **public repos only** — private repos won't be accessible.
- Use `--output json` for machine-readable results.
Guides creation and editing of skills using test-driven development with pressure scenarios and subagents to verify agent compliance.
Runs a structured interview session to sharpen plans or designs, producing ADRs and a glossary as output.
Applies curated color/font themes to slides, docs, and HTML artifacts. Includes 10 preset themes and can generate custom themes on demand.
npx claudepluginhub helloworldsungin/ark-ai-mcp --plugin ark-ai-mcp