Borrowing it
Nothing to install: this file belongs to us-all/mlflow-mcp-server. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/us-all/mlflow-mcp-server/main/CLAUDE.mdgit clone --depth 1 https://github.com/us-all/mlflow-mcp-serverWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/us-all/mlflow-mcp-server/claude-md)<a href="https://agentmods.dev/instructions/us-all/mlflow-mcp-server/claude-md"><img src="https://agentmods.dev/badge/instructions/us-all/mlflow-mcp-server/claude-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.03898 | $0.03898 |
| Opus 5 | $0.01949 | $0.01949 |
| Sonnet 5 | $0.00780 | $0.00780 |
| Haiku 4.5 | $0.00390 | $0.00390 |
Grade A, and why
mlflow-mcp-server CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 141 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md
이 파일은 Claude Code가 이 저장소에서 작업할 때 참고하는 컨텍스트입니다.
프로젝트 개요
@us-all/mlflow-mcp — MLflow REST API를 MCP stdio 서버로 노출하는 TypeScript 구현. 66개 도구 + 4개 워크플로우 프롬프트로 experiments / runs / registered-models / model-versions / logged-models(v3) / traces / assessments 전 영역 커버. 쓰기는 MLFLOW_ALLOW_WRITE=true로 명시 옵트인.
- 타겟 MLflow: 3.5.1+ (v3 traces/assessments REST 사용)
- 런타임: Node 20+, stdio transport (vitest 4 의존성)
- 인증: Bearer 토큰(Databricks PAT) / Basic auth
- 빌드:
pnpm build→dist/index.js
디렉토리
src/
├── index.ts # MCP 도구 66개 + 프롬프트 4개 등록
├── config.ts # 환경변수 로딩
├── client.ts # MlflowClient (fetch 래퍼, /api/ 절대경로 지원, DELETE w/ body)
├── prompts.ts # MCP 워크플로우 프롬프트 4개
└── tools/
├── utils.ts # wrapToolHandler, assertWriteAllowed, sanitize, applyExtractFields
├── experiments.ts (9)
├── runs.ts (17, get-best-run/compare-runs/search-runs-by-tags 포함)
├── registered-models.ts (12)
├── model-versions.ts (9)
├── logged-models.ts (8, MLflow 3 LoggedModel)
├── traces.ts (6, v3 REST + extractFields)
└── assessments.ts (5, v3 REST)
dev/
├── seed.py # demo 실험/런/모델/트레이스 시딩 (idempotent)
└── smoke.mjs # 66 도구 + 4 프롬프트 전수 호출 자동 테스트
docker-compose.yml # mlflow v3.12.0 + seed + mcp 프로필
로컬 검증
docker compose up -d mlflow
docker compose run --rm seed # 실험/런/모델/트레이스까지 모두 시드됨
pnpm build && node dev/smoke.mjs # 76/76 통과 기대
설계 원칙
- Read-only by default: 쓰기 도구는
assertWriteAllowed()로 게이트 - REST 직통: MLflow Python SDK 의존 없음
- 민감정보 마스킹: 에러 메시지에서 token/password 패턴
[REDACTED] - camelCase ↔ snake_case: 도구 핸들러에서 수동 변환
최근 변경사항
- v1.12.7 (2026-06-19): MCP tool annotations 적용 —
@us-all/mcp-toolkit ^1.3.0의inferToolAnnotations를 중앙tool()헬퍼에 추가, 전 도구에 readOnlyHint/destructiveHint/openWorldHint 자동 부여 (override 0). 비파괴 패치. 22/22 test. - v1.12.2 (2026-05-17): 3.12.0 라이브 smoke 회귀 결과 반영. (1)
search-traces.maxResults클라이언트 clamp = 500 (MLflow 3.12+ 서버가 per-page max_results>500을 INVALID_PARAMETER_VALUE로 거부; v1.12.0 changelog 의 "1000-trace 한계 제거" 표현은 부정확이었음 — 실측 cap은 오히려 1000→500). pageToken으로 unbounded 페이지 순회는 정상. (2)list-trace-attachments/get-trace-attachmentdescribe를 "Databricks MLflow only — OSS 서버는 404"로 정정. OSS MLflow 3.12.0 handler 목록에 attachment 라우트 0개 확인. v1.7.0 의 "MLflow 3.9+" 주석은 추측 기반이었음. 76/76 smoke 유지. - v1.12.1 (2026-05-15): 보안 —
pnpm.overrides에 fast-uri ^3.1.2 / hono ^4.12.18 / ip-address ^10.1.1 추가 (CVE-2026-6321/6322, 44455~44459, 42338 transitive 흡수). toolkit ^1.2.2 → ^1.2.3. - v1.12.0 (2026-05-15): MLflow 3.12.0 트래킹 — docker-compose 핀 3.11.1 → 3.12.0.
search-traces는max_results/page_tokenpass-through라 코드 변경 0줄로 진행. attachment 도구 회귀 검증은 v1.12.2 에서 수행 → 위 항목 참조.@us-all/mcp-toolkit^1.2.1 → ^1.2.2 dep 핀 동반 cascade. - v1.11.2 (2026-05-06): MCP Server Registry 발행 —
mcpName: "io.github.us-all/mlflow"추가 + 루트server.json(MLFLOW_TRACKING_URI required + token/basic-auth optional 메타데이터). 코드 변경 0줄. - v1.11.1 (2026-05-05):
@us-all/mcp-toolkit ^1.2.1핀 업데이트 — 자동 cascade. 코드 변경 0줄. - v1.11.0 (2026-05-05): Apps SDK UI 카드 —
compare-runs도구 결과를_meta["openai/outputTemplate"]통해 ChatGPT/Apps SDK 클라이언트에서 카드로 렌더 (run summary + metric/param 테이블, differing_params 하이라이트). 새 리소스ui://widget/compare-runs.html(text/html+skybridge). Claude 클라이언트는_meta무시 — non-breaking. 빌드 시src/ui/*.html을dist/ui/로 자동 복사. - v1.10.0 (2026-05-05):
startMcpServer채택 — toolkit v1.2.0의 런타임 헬퍼로 stdio 부트스트랩을 1줄로 교체.MCP_TRANSPORT=http로 Streamable HTTP transport 옵트인 가능 (기본 stdio). Bearer 인증,/health엔드포인트. 기존 stdio 사용자 영향 0. - v1.9.1 (2026-05-05):
@us-all/mcp-toolkit ^1.2.0핀 업데이트 — 자동 cascade. 코드 변경 0줄. - v1.9.0 (2026-05-04): MCP Prompt
analyze-failed-traces추가 — 기존debug-failed-traces(one-shot 조사)와 보완. 현재 윈도우 vs 직전 윈도우 비교로 실패율 추세, top N 실패 패턴 by 임팩트, 새 regression 식별. MLflow 3 GenAI 트레이싱 흐름 + 가시성 sprint Week 2 P0 항목. - v1.8.5 (2026-05-03):
@us-all/mcp-toolkit ^1.1.0채택 +aggregate()헬퍼로summarize-experiment마이그레이션. caveats 라벨 텍스트 변경 (get-experiment failed:→getExperiment failed:,search-runs failed:→searchRuns failed:).summarize-run은Promise.all+ per-fetcher.catch()패턴이라 마이그레이션 대상 외 — 그대로 유지. - v1.8.4 (2026-05-03):
@us-all/mcp-toolkit ^1.0.0핀 업데이트. toolkit API freeze (semver 1.x 보장 시작) — 코드 변경 0줄, 20/20 테스트 통과. - v1.8.3 (2026-05-03): Wave 6 —
summarize-run의 metricHistory 각 포인트에서 중복key필드 제거(4k-point 기준 ~100KB 절감).summary.artifactsIncluded가 fetch 실패 시({error:...})에도 truthy였던 문제 수정 — 실제 fetch 성공 여부 반영. - v1.8.2 (2026-05-03):
summarize-experiment의 topRuns metrics 항목에서 step/timestamp default drop (key/value만). caller-supplied extractFields 우선. - v1.8.1 (2026-05-02):
summarize-runmetric history가 MLflow 3.x에서 비어있던 문제 —/metrics/get-history가max_results없이 호출되면 next_page_token만 반환. 25000(문서상 최대) 명시. - v1.8.0 (2026-05-02):
summarize-experiment어그리게이션 — experiment + topN runs (metric/start_time 정렬) + metric stats(min/max/mean) 1 call. 3-5 round-trip 대체. - v1.7.0 (2026-05-02): Wave 3 Resources 확장 —
mlflow://run/{runId}/artifacts,mlflow://experiment/{expId}/runs,mlflow://registered-model/{name}/versions추가. 또한 trace attachment 도구 2개(list-trace-attachments,get-trace-attachment, MLflow 3.9+). - v1.6.2 (2026-05-02): Wave 1 — describe trim 13, 의존성 bumps, fat-read 3개에 default extractFields.
- v1.6.1 (2026-05-02):
@us-all/mcp-toolkit ^0.2.0채택 — 로컬sanitize/wrapToolHandler본문 제거,createWrapToolHandlerfactory로 위임.redactionPatterns: [/basic\s+\S+/i]+errorExtractors(WriteBlockedError → passthrough, MlflowError → structured{status, body})만 명시. utils.ts 87→56 lines. - v1.6.0 (2026-05-01):
@us-all/mcp-toolkit ^0.1.0으로 마이그레이션 —tool-registry.ts와applyExtractFields를 toolkit에서 import. utils.ts의 inline 구현 제거. 단위 테스트(tool-registry, extract-fields)는 toolkit이 owns. ~170 lines 코드 절감. - v1.5.1:
pnpm token-stats스크립트 + CI TOKEN_BUDGET=12000 가드 추가. - v1.5.0:
summarize-run어그리게이션 도구 — run info + (opt) metric history per key + (opt) artifacts. 3-5 round-trips → 1 call. - v1.4.0: MCP Resources (
mlflow://URI scheme) 6개 — run, experiment, experiment-by-name, registered-model, model-version, trace. - v1.3.1:
extractFieldsauto-apply viawrapToolHandler. experiments / runs 핵심 read 스키마에 명시적 선언. - v1.3.0: 카테고리 ENV 토글(
MLFLOW_TOOLS/MLFLOW_DISABLE) 8 카테고리 +search-tools메타툴. - v1.2.0: Webhooks 6개 + Prompt Optimization 5개 도구 추가 (도구 66→77).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 141 lines · 3,898 tokens per session scan A 310a7af14afe
mlflow-mcp-server CLAUDE.md is an instructions file published in the GitHub repository us-all/mlflow-mcp-server (1 stars, last pushed 1mo ago), licensed MIT. It adds 3,898 tokens to every session, about $0.0195 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
codex AGENTS.md
AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.
vscode buildNext.instructions.md
Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).
next.js AGENTS.md
AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.
langchain AGENTS.md
AGENTS.md instructions for langchain-ai/langchain, covering global development guidelines for the langchain monorepo, corridor security analysis, project architecture and context, monorepo structure and development tools & commands.
vscode oss-third-party-notices.instructions.md
Instructions for microsoft/vscode, covering vs code oss third-party-notices pipeline, architecture, pipeline flow in ci, applying the notice (cutover) and fallback chain (never fail the build).
spec-kit AGENTS.md
AGENTS.md instructions for github/spec-kit, covering agents.md, about spec kit and specify, quickstart — add a new integration in 5 steps, integration architecture and integrationmanifest — file tracking.