| name | Docs Research |
| description | Use MCP Central Docs server to search, ask, and retrieve documentation (FastMCP, standards, packaging). |
Docs Research Skill
Description: Fleet documentation research via MCP Central Docs server. Covers doc set indexing, semantic search across documentation repos, cross-repository queries, fleet standards navigation, and autonomous multi-doc research workflows.
Trigger Phrases
- "Find docs about [topic]"
- "Search the fleet documentation for [concept]"
- "How do I [task] according to fleet standards?"
- "What does the standard say about [pattern]?"
- "Show me the documentation for [tool/system]"
- "Research best practices for [technology]"
- "Summarize all standards related to [topic]"
- "What changed in the latest version of [doc]?"
Tools
search_docs(query, limit) — Semantic search over indexed documentation. Use for "find docs about X", "where is Y documented". Returns ranked results with snippet and relative_path.
ask_docs(question) — Get a synthesized answer from the docs (requires client sampling). Use for questions that need synthesis across multiple doc files.
get_document(relative_path) — Fetch full file content after you have a path from search_docs. Always prefer this for reading entire standards or guides.
agentic_doc_workflow(workflow_prompt) — Autonomous multi-step research. Pass a goal string like "Research all onboarding docs and create a summary". Chains search → get_document → ask_docs.
reindex_docs() — Force re-indexing after doc changes. Run when new docs are added or existing docs are updated.
chunk_stats() — Index health: total chunks, doc count, embedding model, last indexed timestamp.
start_webapp() — Start the Docs webapp (backend + frontend + open browser).
Workflow
- Search first: Prefer
search_docs with a clear, specific query. Use limit to control result breadth (default 10, increase for research surveys).
- Deep read: Use
get_document(relative_path) when a snippet isn't enough. Always include filename or relative_path from results for traceability.
- Synthesis: Use
ask_docs when the user wants a direct answer and the client supports sampling. Best for cross-document questions.
- Broad research: Use
agentic_doc_workflow or the research_workflow prompt for comprehensive multi-topic research tasks.
- Index maintenance: Run
reindex_docs() after any doc changes. Check chunk_stats() to verify index health.
Citing Sources
Always include filename or relative_path from results so the user can open or re-fetch the doc. Format: [filename](relative_path) for clickable references.
Indexed Content
The docs index covers: FastMCP standards, MCPB packaging, fleet workflow rules, SOTA patterns, port allocation registry, tool registration protocols, docstring standards, architectural guidelines, PowerShell conventions, and per-repo AGENTS.md files.
Examples
- "Find the standard for MCP tool registration." →
search_docs(query="MCP tool registration", limit=5) → get_document(relative_path="standards/rules/mcp_registration.md")
- "What are the fleet port allocation rules?" →
ask_docs(question="What port range is reserved for fleet webapps and which ports are forbidden?")
- "Research all onboarding docs." →
agentic_doc_workflow(workflow_prompt="Find all onboarding and quickstart documents and create a summary guide")