mcp-name: io.github.nipunkhanderia/golden-dataset-mcp
golden-dataset-mcp
An MCP server wrapping golden-dataset-studio — version-controlled golden dataset management and semantic evaluation for RAG/LLM pipelines.
This is a thin protocol layer over the existing golden_dataset library (DatasetStore, Evaluator). It does not reimplement any logic — it exposes the library's existing Python API as MCP tools so an agent (Claude Desktop, Claude Code, or any MCP client) can manage golden datasets conversationally.
No LLM API key required. Evaluation uses TF-IDF cosine similarity (scikit-learn), not an LLM call.
Why a separate package from golden-dataset-studio?
golden-dataset-studio is a CLI tool — designed for a human typing golden add, golden commit, etc. in a terminal. golden-dataset-mcp exposes the same underlying operations as MCP tools so an LLM agent can drive them programmatically, e.g. as part of an automated RAG evaluation pipeline. Keeping them as separate PyPI packages means CLI users aren't forced to pull in fastmcp as a dependency, and MCP users get a clean, protocol-focused package.
Tools
| Tool | What it does |
|---|---|
init_dataset | Initialise a new dataset at a given path |
add_entry | Add a question/answer pair to the working tree |
update_entry | Edit fields of an existing working-tree entry |
delete_entry | Remove an entry from the working tree |
list_entries | List working-tree or committed-version entries |
commit_version | Snapshot the working tree as a new immutable version |
diff_versions | Show entries added/removed/changed between two versions |
evaluate_answers | Score actual answers against a version via TF-IDF cosine similarity |
dataset_status | Show current version, working tree size, and version history |
Design: every tool takes an explicit dataset_path
Unlike the CLI (which operates on the current working directory), every tool here requires an explicit dataset_path parameter. This keeps the server fully stateless between calls — no hidden "current dataset" session state to lose track of, and safe for one server instance to manage multiple datasets or serve multiple concurrent clients.
Installation
pip install golden-dataset-mcp
This pulls in golden-dataset-studio and scikit-learn automatically as dependencies.
Usage with Claude Desktop / Claude Code
{
"mcpServers": {
"golden-dataset": {
"command": "golden-dataset-mcp"
}
}
}
No environment variables needed — no API key, no config.
Example flow
1. init_dataset(dataset_path="./my-rag-eval", name="support-bot-eval")
2. add_entry(dataset_path="./my-rag-eval", question="...", answer="...")
[repeat for each golden Q&A pair]
3. commit_version(dataset_path="./my-rag-eval", description="initial 50 questions")
4. [run your RAG pipeline, collect actual answers]
5. evaluate_answers(dataset_path="./my-rag-eval", actual_answers=[...])
-> avg_semantic_similarity, per-entry scores, pass/fail
As your RAG pipeline changes over time, commit_version again after edits and use diff_versions to see exactly what changed in your golden set between releases.
Relationship to the underlying library
golden-dataset-studio | golden-dataset-mcp | |
|---|---|---|
| Interface | CLI (golden ...) | MCP tools |
| Driven by | A human typing commands | An LLM agent / MCP client |
| Path handling | Current working directory | Explicit dataset_path per call |
| Dependency direction | — | Depends on golden-dataset-studio |
If you want the human-driven CLI, use golden-dataset-studio directly. If you want an agent to drive it, use this package.
Development
git clone https://github.com/nipunkhanderia/golden-dataset-mcp
cd golden-dataset-mcp
pip install -e ".[dev]"
pytest -v
Validate the MCP-facing contract:
npx @modelcontextprotocol/inspector golden-dataset-mcp
Limitations
evaluate_answersuses TF-IDF cosine similarity, which captures lexical overlap better than deep semantic meaning. For embedding-based or RAGAS-style metrics, call the underlying library'sEvaluator.ragas_evaluate()directly (requirespip install "golden-dataset-studio[ragas]"— not exposed as an MCP tool in this version).- Very short or stop-word-only answers will raise an error. scikit-learn's TF-IDF vectorizer raises
ValueError: empty vocabularyon inputs like a bare"4"or"the a an". Avoid single-token golden answers, or expectevaluate_answersto fail on them. - All state is filesystem-backed JSON/JSONL under
<dataset_path>/.golden_dataset/; this server does no remote storage or syncing.
License
MIT