멀티 LLM 오케스트레이션 데스크톱 앱 및 웹 서버. 브라우저 기반 오케스트레이션 UI, CLI 세션 관리, MCP 호스트, 채팅 및 프로젝트 실행 기능 제공.
Fleet's MCP server has critical definition gaps. All 4 tools have schemas with proper JSON structure and type declarations, but descriptions are minimal (10-20 chars) and lack actionable context. Tool names are verb-based (read_file, list_directory, grep, echo) but descriptions are in Korean and extremely terse, providing no guidance on WHEN to use each tool or what downstream operations expect. Parameters lack descriptions entirely in most cases (e.g., 'path' in read_file has a description, but 'pattern' in grep lacks context on regex syntax). No output schemas are documented. Error handling is absent, no guidance on recovery, retryability, or what to do when a file is not found. The echo tool appears to be test-only and should not be exposed in production.
입력을 그대로 반환한다(테스트용).
워크스페이스 내 파일 내용을 정규식으로 검색한다. 결과는 "상대경로:줄번호:내용".
워크스페이스 내 디렉터리 항목을 나열한다(path 생략 시 루트). 디렉터리는 / 접미사.
워크스페이스 내 텍스트 파일을 읽는다. path 는 워크스페이스 루트 기준 상대경로.
Descriptions are under 20 characters and in Korean, providing no actionable context for LLM tool selection. Examples: 'read_file' = '워크스페이스 내 텍스트 파일을 읽는다' (18 chars, untranslated). LLMs cannot determine WHEN to use this tool vs alternatives.
No output schemas documented for any tool. LLMs cannot plan downstream operations or extract required fields. E.g., does list_directory return {items: [{name, type}]} or {files: [], dirs: []}? Does grep return {matches: [{file, line, content}]} or raw text?
Parameter descriptions missing or minimal. 'pattern' in grep has no description of regex syntax expectations. 'path' in list_directory is optional but no description explains what happens when omitted (defaults to root?).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2025-06-18+ | v2 |
No error handling guidance. What happens if read_file targets a binary file, a missing file, or a path outside the workspace? No recovery hints, retryability classification, or actionable error messages.
echo tool is test-only (source: e2e/fixtures/mock-mcp-server.mjs) and should not be exposed in production. Pollutes the tool namespace and confuses LLM selection logic.