Borrowing it
Nothing to install: this file belongs to berkayildi/mcp-llm-eval. 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/berkayildi/mcp-llm-eval/main/CLAUDE.mdgit clone --depth 1 https://github.com/berkayildi/mcp-llm-evalWrote 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/berkayildi/mcp-llm-eval/claude-md)<a href="https://agentmods.dev/instructions/berkayildi/mcp-llm-eval/claude-md"><img src="https://agentmods.dev/badge/instructions/berkayildi/mcp-llm-eval/claude-md/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/instructions/berkayildi/mcp-llm-eval/claude-md"><img src="https://agentmods.dev/badge/instructions/berkayildi/mcp-llm-eval/claude-md.svg" alt="Reviewed on agentmods" width="80" 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 | $0.02639 | $0.02639 |
| Opus 5.5 | $0.01056 | $0.01056 |
| Sonnet 5.5 | $0.00528 | $0.00528 |
| Haiku 4.5 | $0.00264 | $0.00264 |
Grade A, and why
mcp-llm-eval 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 2d 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 — 223 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CLAUDE.md — mcp-llm-eval
This file helps Claude Code understand the project structure and conventions for future sessions.
Project overview
mcp-llm-eval is a local MCP (Model Context Protocol) server written in Python. It packages LLM evaluation gates as reusable CI/CD primitives — load a dataset, run models, score with an LLM-as-judge, and check quality thresholds, all exposed as MCP tools that AI agents can call.
Transport: stdio (standard input/output).
Directory structure
mcp-llm-eval/
├── src/
│ └── mcp_llm_eval/
│ ├── __init__.py # Package version (__version__ = "0.1.0")
│ ├── server.py # MCP server + tool registration (6 tools) + entry point routing
│ ├── engine.py # Eval engine: dataset loader, LLM runner, judge, threshold checker
│ ├── cli.py # CLI argument parsing and subcommand routing (run, check, compare, comment)
│ ├── config.py # .eval-gate.yml loader and validator
│ ├── comparison.py # compare_runs logic: regression detection with tolerance
│ ├── formatter.py # PR comment markdown generator
│ ├── providers/
│ │ ├── __init__.py
│ │ ├── anthropic.py # Streaming runner for Anthropic (messages.stream, TTFT capture)
│ │ ├── openai.py # Streaming runner for OpenAI (chat.completions.create stream)
│ │ └── google.py # Streaming runner for Google GenAI (generate_content_stream)
│ ├── judge.py # LLM-as-judge scorer (faithfulness + relevance, 0-1 scale)
│ ├── retrieval.py # RetrievalAdapter protocol + BM25Adapter
│ ├── retrieval_metrics.py # IR metrics (recall@k, precision@k, MRR, nDCG@k)
│ ├── embeddings.py # OpenAIEmbeddingAdapter + GoogleEmbeddingAdapter (v0.7.0)
│ └── types.py # Shared dataclasses: EvalEntry, EvalResult, RunSummary, ThresholdConfig
├── tests/
│ ├── __init__.py
│ ├── fixtures/
│ │ └── sample_dataset.json # 3 sample eval entries (one per category: factual, reasoning, summarization)
│ ├── test_server.py # MCP tool integration tests (6 tools)
│ ├── test_engine.py # Eval engine unit tests
│ ├── test_providers.py # Provider runner tests (mock API calls)
│ ├── test_judge.py # Judge scoring tests (mock OpenAI)
│ ├── test_types.py # Type validation tests
│ ├── test_cli.py # CLI argument parsing, subcommand routing, exit codes
│ ├── test_config.py # YAML loading, validation, defaults, error handling
│ ├── test_comparison.py # Regression detection, tolerance math, edge cases
│ └── test_formatter.py # Markdown generation, with/without comparison and thresholds
├── pyproject.toml # Build config (hatchling), deps, pytest settings
├── Makefile # setup / build / start / test / clean
├── README.md # User-facing docs
├── CLAUDE.md # This file
├── LICENSE # MIT
├── CHANGELOG.md # Release Please manages this
├── release-please-config.json
├── .release-please-manifest.json
└── .github/
└── workflows/
└── release.yml # Release Please + PyPI OIDC publish
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.
- 2d ago Changed · -14 tokens per session 1e9ae69cafed
- 2mo ago First seen · 223 lines · 2,653 tokens per session scan A aa6994abcd1c
mcp-llm-eval CLAUDE.md is an instructions file published in the GitHub repository berkayildi/mcp-llm-eval (0 stars, last pushed 3d ago), licensed MIT. It adds 2,639 tokens to every session, about $0.0106 per session on Opus 5.5. 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.
mcp-llm-eval CLAUDE.md compared with instructions from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| mcp-llm-eval CLAUDE.mdberkayildi/mcp-llm-eval | Instructions file | 0 | 3d ago | MIT | A | 2,639 | Claude Code |
| vscode buildNext.instructions.mdmicrosoft/vscode | Instructions file | 195k | yesterday | MIT | A | 9,719 | GitHub Copilot |
| next.js AGENTS.mdvercel/next.js | Instructions file | 143k | yesterday | MIT | A | 7,461 | Codex, OpenCode |
| langchain AGENTS.mdlangchain-ai/langchain | Instructions file | 149k | yesterday | MIT | A | 4,680 | Codex, OpenCode |
| vscode oss-third-party-notices.instructions.mdmicrosoft/vscode | Instructions file | 195k | yesterday | MIT | A | 5,516 | GitHub Copilot |
| vscode copilot-instructions.mdmicrosoft/vscode | Instructions file | 195k | yesterday | MIT | A | 3,093 | GitHub Copilot |
| deepseek-harness AGENTS.mddeepseek-ai/deepseek-harness | Instructions file | 274k | 2d ago | MIT | A | 4,313 | Codex, OpenCode |