AMED研究项目检索数据库

AMED Research Database

A searchable database and API for research proposals accepted by AMED (Japan Agency for Medical Research and Development), designed for analysis by humans and LLMs alike.

What's Inside

• ~3,800 projects from 2015-2025, scraped from AMED's public listings
• 3,000+ researchers, 700+ institutions, 200+ programs
• Pre-built SQLite database included (database/amed.db) - ready to query immediately
• REST API with search, analytics, and LLM-optimized endpoints
• MCP server for direct integration with Claude and other LLM tools
• PDF summary report with charts and analysis

Quick Start

pip install -r requirements.txt

# Start the API server
uvicorn api.main:app --reload
# → http://localhost:8000/docs

The MCP server needs a separate environment, because mcp requires newer starlette and pydantic releases than the API stack pins:

pip install -r requirements-mcp.txt
python mcp_server.py

Docker

docker compose up
# → http://localhost:8000/docs

API Endpoints

Search

• GET /api/projects/search?q=がん&year=2024 - Search projects (supports format=compact)
• POST /api/projects/bulk - Bulk query multiple searches

Statistics

• GET /api/stats - Overview statistics
• GET /api/metadata - Database metadata for LLM context
• GET /api/suggestions - Suggested queries

Analytics (/api/analytics/)

• trends/by-year - Project count trends
• trends/keywords?keywords=AI,がん,ゲノム - Keyword frequency over time
• institutions/top?top_n=15 - Top institutions
• institutions/compare?names=東京大学,京都大学 - Side-by-side comparison
• institutions/collaborators?institution=東京大学 - Co-participating institutions
• programs/top - Top research programs
• researchers/profile?name=田中 - Researcher lookup
• researchers/young-ratio - Young researcher program stats

MCP Server

For use with Claude Code, Claude Desktop, or any MCP-compatible client. Install requirements-mcp.txt first (see Quick Start), then:

{
  "mcpServers": {
    "amed": {
      "command": "python",
      "args": ["mcp_server.py"],
      "cwd": "/path/to/amed-research-db"
    }
  }
}

Tools: search_projects, get_stats, top_institutions, compare_institutions, keyword_trends, researcher_profile, find_collaborators

Project Structure

amed-research-db/
├── api/
│   ├── main.py          # FastAPI app with search, stats, metadata
│   └── analytics.py     # Analytics endpoints (trends, comparisons)
├── database/
│   ├── models.py        # SQLAlchemy models (Project, Researcher, etc.)
│   ├── import_all.py    # Batch data importer
│   └── amed.db          # Pre-built SQLite database
├── scraper/
│   ├── scrape_all.py    # Full scraper: link collection + progress tracking
│   ├── parser.py        # HTML parser (header-based column detection)
│   ├── check_new.py     # Detect newly published announcements
│   └── find_missing.py  # Re-scrape pages missing from the database
├── tests/               # 65 tests (parser + API)
├── mcp_server.py        # MCP server for LLM integration
├── requirements.txt     # API, scraper, tests
├── requirements-mcp.txt # MCP server (separate environment)
├── AMED_Report.pdf      # Summary report with charts
├── Dockerfile
└── docker-compose.yml

Rebuilding the Database

If you want to re-scrape and rebuild from scratch:

# 1. Collect links and scrape all project pages
#    (takes ~30 min, respectful rate limiting; resumes from saved progress)
python scraper/scrape_all.py

# 2. Import into database
python database/import_all.py

Tech Stack

• Python 3.10+, FastAPI, SQLAlchemy, SQLite
• BeautifulSoup4 (scraping), pytest (testing)
• MCP (Model Context Protocol) for LLM integration
• reportlab + matplotlib (PDF report)

Data Source

All data is from AMED's publicly available accepted proposal listings at https://www.amed.go.jp/koubo/saitaku_index.html

Citation

Each release is archived on Zenodo. Cite the concept DOI 10.5281/zenodo.22882293, which always resolves to the latest version, or the DOI of the specific version you used (v1.0.0 is 10.5281/zenodo.22882294).

Ohta, T. (2026). AMED Research Database: a searchable database, REST API, and MCP server for AMED-funded research projects (v1.0.0) [Software]. Zenodo. https://doi.org/10.5281/zenodo.22882293

See CITATION.cff, or GitHub's "Cite this repository" button, for a machine-readable version.

License

• Code — MIT (LICENSE)
• Data and report — CC BY 4.0: database/amed.db and AMED_Report.pdf (LICENSE-DATA)

The CC BY 4.0 terms cover the compilation and the derived analyses, not the upstream listings themselves. Check AMED's own terms of use before redistributing the source material.