MCP server exposing Koog framework documentation to any MCP-capable client
Three well-named tools with clear, action-oriented descriptions (list_, search_, get_). All tools have documented input schemas with type constraints and descriptions. Descriptions are 150 - 250 chars, within the 10 - 1024 baseline. Parameters include constraints (min/max, positive, integer). However, output schemas are not formally documented, responses are inferred from code (JSON content for list/search, markdown text for get_skill). Error handling in get_skill returns structured error responses but lacks recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all three being read-only operations. Resource template is well-structured but adds complexity without clear agent benefit.
Fetch the full SKILL.md content for a named skill. Set include_references=true to inline every linked reference file. Always read the returned skill in full before acting on it — it contains step-by-step procedures, code examples, and API references you must follow.
List all available Koog framework skills. Each skill is a self-contained guide for a specific Koog feature (e.g. agents, tools, memory, RAG, streaming, MCP integration). Use this when the user asks an open-ended Koog question and you want to discover what curated guidance exists. Prefer search_skills when you have a specific query.
Full-text search the Koog framework skills library by keyword or natural-language query. Returns ranked matches with a short snippet. Call this BEFORE answering any non-trivial Koog question — the curated skills carry the canonical JetBrains guidance. Then call get_skill with the top result to read its full content.
Output schemas not formally documented. Responses are inferred from code (JSON for list/search, markdown text for get_skill). LLMs cannot plan downstream tool calls or extract structured data without explicit output schema declarations.
No tool annotations despite all three tools being read-only. Adding readOnlyHint=true would signal to agents that these calls are safe to retry and have no side effects.
Error handling in get_skill returns 'isError: true' but lacks recovery guidance. Should suggest 'Call list_skills() to see available skill names' when skill not found.
Resource template (skill://{name}) duplicates get_skill functionality without clear added value. Agents must reason about whether to use the tool or resource URI, unnecessary cognitive load.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
search_skills returns 'hits' array but structure is not documented. Does each hit include name, description, snippet, score? LLMs must infer from code.