Public remote-MCP server serving KISA's curated vibe-coding experience: knowledge, ready-made skills, leveling-up routes and recommendations.
Seven read-only tools with clear verb-noun naming (search_, get_, list_, next_). All have descriptions (194 - 350 chars, within baseline 34 - 392). Input schemas present with types and descriptions. Output schemas documented in docstrings. Key strength: descriptions explain WHEN to call each tool and what it returns. Weakness: no parameter constraints (enums, ranges), no error recovery guidance, no tool annotations (readOnlyHint missing despite all tools being read-only). Composition is sound, tools chain logically (search → get_item, onboard → next_step). No security issues (read-only, no secrets). Moderate gaps prevent A-grade.
Get the FULL content of any catalog item by id — knowledge, tool, route or skill. search_knowledge and list_recommended give only a title and a short snippet. Once you know which item you want, call get_item(id) for the complete guide body. Works for every type: knowledge and tool guides (otherwise unreachable in full), skills (same body as get_skill), and routes (overview + step list — walk them with next_step).
Get the full pack for one skill: install steps, config and checklist.
List recommended items: the best entry points when the user is unsure where to start.
List installable skills — ready-made packs (configs, steps, checklists). Use this to see what the agent can install for the user, then call get_skill(id) for the full pack. The base skill (tailored-install) comes first: it is step 0 of every install — always apply its discipline before installing anything.
Get one route step's materials and action. See onboard() for context.
No tool annotations (readOnlyHint) despite all tools being read-only. LLMs cannot infer safety from names alone; explicit annotations prevent accidental misuse.
Parameter 'limit' in search_knowledge lacks min/max constraints. Unbounded integers let LLMs pass absurd values (limit=999999). Should enforce 1 - 50 range.
Error handling in get_item and get_skill raises ToolError but does not suggest recovery steps. 'Item not found' should include 'Try search_knowledge() or list_skills().'
Output schemas documented in docstrings but not in formal JSON Schema. FastMCP should register schemas in tool definitions for machine-readability.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Get a step-by-step leveling-up route for a goal: what to read and install, in order. Call this when the user has a concrete goal ("connect Tavily", "harden my VPS", "go from zero to shipping"). Returns the whole route with an ordered list of step ids; then call next_step(step_id) for each step's materials. Onboarding is stateless — you hold the progress, the server does not.
Search KISA's curated notes, tool guides and skills by a natural-language query. Call this whenever you need context on a vibe-coding topic (an agent setup, a tool, a technique, a pattern KISA uses) or when the user wants a skill on a specific topic. Returns the most relevant fragments so you can drop them straight into your context.
list_recommended and list_skills return no pagination info (total count, next_cursor). If result sets grow, LLMs cannot fetch additional pages.