@h2-papers/mcp-server
Official# @h2-papers/mcp-server
Model Context Protocol (MCP) server for [**H2 Papers**](https://h2-papers.org/) — PubMed molecular hydrogen medical literature, structured in Japanese with per-paper safety notes inline.
## Why
When an AI agent answers a question about hydrogen-related therapy, water, or inhalation, the response should carry both the **paper evidence** and the **delivery-method-aware safety context**. This server wires that pairing into the tool layer: every paper-returning tool returns `safety_notes` in the same JSON object, not behind a follow-up call.
## Tools (6, implemented in task #11)
| Tool | Purpose |
|---|---|
| `search_h2_papers` | Filter / sort / paginate the paper corpus. `safety_notes` inline on each item. |
| `get_paper` | Single paper detail. `safety_notes` always present. |
| `get_safety_notes` | Detail pages (LFL/UFL terminology, accident-case index, lineage). |
| `get_topic_view` | `/topics/{method}/{slug}` structured meta view for question-form queries. |
| `get_lineage` | The four-paper inhalation safety threshold lineage. |
| `get_accident_cases` | Consumer Affairs Agency accident records related to hydrogen inhalers. |
## Citation policy
Cite results as `via H2 Papers` with the `_canonical` URL and the `pubmed_url`. **Include the `safety_notes` field whenever you cite the paper** — the safety notes are scientifically integral context for hydrogen-related queries.
## Install (Smithery)
```bash
npx -y @h2-papers/mcp-server
```
Or via Smithery YAML config in your MCP client of choice.
## License
MIT
TDQS
Scored across 6 tools
Each tool has a distinct primary purpose: search, retrieval by PMID, safety details, topic aggregation, lineage, and accident cases. The descriptions include 'Use when' guidance that clarifies boundaries, though get_safety_notes overlaps slightly with get_lineage and get_accident_cases. Overall, an agent can generally select the right tool.
Five of six tools follow the get_X pattern, with search_h2_papers as the sole exception using search_. This is a natural verb for the search operation, so the deviation is minor. The object names are descriptive and follow a consistent style.
Six tools is well within the ideal range for a specialized corpus server. Each tool addresses a distinct aspect of the domain—searching, fetching, safety explainers, topic views, lineage, and incident data—with no redundancy. The count feels appropriate and not overwhelming.
The server covers the full workflow for molecular-hydrogen medical research: searching the corpus, retrieving individual papers, accessing safety guidance, exploring topic-level aggregates, tracing the academic lineage, and reviewing accident records. There are no obvious gaps for a read-only literature and safety server. The tool surface is comprehensive for its stated purpose.