feat: add PubMed Search MCP server (31 literature research tools)

[u9401066]

Description

Add PubMed Search MCP, a professional literature research assistant MCP server with 31 tools for comprehensive biomedical research.

Changes Made

• MCP Server: Complete MCP server via uvx pubmed-search-mcp
• Configuration: 31 tools in data/remote_tools/pubmed_search_mcp_tools.json

File Structure (in pubmed-search-mcp repo)

src/pubmed_search/
 mcp/
    server.py              # MCP server entry point
    tools/
        discovery.py       # search, related, citing, details, metrics
        strategy.py        # generate_queries, analyze_query
        pico.py            # parse_pico
        citation_tree.py   # build_citation_tree, suggest_citation_tree
        europe_pmc.py      # get_fulltext, get_text_mined_terms
        export.py          # prepare_export
        ncbi_extended.py   # gene, compound, clinvar tools
        openurl.py         # institutional access tools
        vision_search.py   # figure analysis tools
        unified.py         # unified_search
 sources/                   # API clients
    europe_pmc.py          # Europe PMC client
    semantic_scholar.py    # Semantic Scholar client
    openalex.py            # OpenAlex client
    core.py                # CORE client
    ncbi_extended.py       # Gene, PubChem, ClinVar
 session.py                 # Session management

Tools Included (31 tools)

Core Search & Strategy (4)

Tool	File	Description
pubmed_unified_search	unified.py	Multi-source search
pubmed_analyze_search_query	unified.py	Query analysis
pubmed_parse_pico	pico.py	PICO parsing
pubmed_generate_search_queries	strategy.py	MeSH expansion

Discovery & Citation (7)

Tool	File	Description
pubmed_find_related_articles	discovery.py	Similar articles
pubmed_find_citing_articles	discovery.py	Forward citation
pubmed_get_article_references	discovery.py	Backward citation
pubmed_fetch_article_details	discovery.py	Article metadata
pubmed_get_citation_metrics	discovery.py	iCite RCR metrics
pubmed_build_citation_tree	citation_tree.py	Citation network
pubmed_suggest_citation_tree	citation_tree.py	Tree evaluation

Full Text & Export (3)

Tool	File	Description
pubmed_get_fulltext	europe_pmc.py	Structured full text
pubmed_get_text_mined_terms	europe_pmc.py	Text-mined annotations
pubmed_prepare_export	export.py	RIS/BibTeX/CSV export

NCBI Extended (7)

Tool	File	Description
pubmed_search_gene	ncbi_extended.py	Gene search
pubmed_get_gene_details	ncbi_extended.py	Gene details
pubmed_get_gene_literature	ncbi_extended.py	Gene literature
pubmed_search_compound	ncbi_extended.py	PubChem search
pubmed_get_compound_details	ncbi_extended.py	Compound details
pubmed_get_compound_literature	ncbi_extended.py	Compound literature
pubmed_search_clinvar	ncbi_extended.py	ClinVar variants

Vision & Institutional (6)

Tool	File	Description
pubmed_analyze_figure_for_search	vision_search.py	Figure analysis
pubmed_reverse_image_search	vision_search.py	Reverse image search
pubmed_configure_institutional_access	openurl.py	OpenURL config
pubmed_get_institutional_link	openurl.py	Access link
pubmed_list_resolver_presets	openurl.py	Resolver presets
pubmed_test_institutional_access	openurl.py	Test access

Session Management (4)

Tool	File	Description
pubmed_get_session_pmids	session_tools.py	Cached PMIDs
pubmed_list_search_history	session_tools.py	Search history
pubmed_get_cached_article	session_tools.py	Cached retrieval
pubmed_get_session_summary	session_tools.py	Session summary

Additional AI Agent Support

This tool also includes 22 Claude Skill files (.claude/skills/) that provide:

• Step-by-step workflow guidance for AI agents
• Decision trees for search strategy selection
• Code examples ready for immediate use

Available Skills:

• pubmed-quick-search - Basic PubMed search with advanced filters
• pubmed-systematic-search - MeSH expansion, comprehensive search
• pubmed-pico-search - PICO clinical question decomposition
• pubmed-paper-exploration - Citation tree, related articles
• pubmed-gene-drug-research - Gene, PubChem, ClinVar integration
• pubmed-fulltext-access - Europe PMC, CORE full text retrieval
• pubmed-export-citations - RIS, BibTeX, CSV export
• pubmed-multi-source-search - Cross-database search strategy
• pubmed-mcp-tools-reference - Complete tools reference

Note: Skill files are Claude Code-specific (.claude/skills/*/SKILL.md). ToolUniverse uses JSON configs for tool integration, which is already provided in this PR. If TU adds similar workflow guidance support in the future, these skills can be adapted.

Test Results

Check	Status	Details
pytest	Pass	565 passed, 13 skipped (578 total)
ruff lint	Pass	All checks passed
bandit security	Pass	No high severity issues

Server Information

• Protocol: MCP (Model Context Protocol)
• Transport: stdio / SSE / Streamable HTTP
• Installation: uvx pubmed-search-mcp or pip install pubmed-search-mcp
• Repository: https://github.com/u9401066/pubmed-search-mcp
• PyPI: https://pypi.org/project/pubmed-search-mcp/
• License: Apache-2.0

Checklist

• JSON config valid (31 tools)
• Published on PyPI
• Documentation complete
• Tests passing (565/578)
• Security scan passed
• Claude Skills included (22 workflow guides)

[u9401066]

Hi maintainers,

Thank you for reviewing this PR! I wanted to share some context about the design decisions:

Regarding local vs. remote MCP integration:
I'm not entirely sure if this MCP server should be built-in as a local tool in ToolUniverse, given:

• It has several dependencies (biopython, requests, pylatexenc, etc.)
• There might be some overlap with existing search functionality in TU

Development approach:
This project was developed collaboratively with GitHub Copilot + Claude Opus, following:

• Domain-Driven Design (DDD) architecture
• Comprehensive automated checks (pytest, ruff lint, bandit security)
• 90%+ test coverage

Key features designed for AI agents:

• Auto-correction: ESpell integration for typo fixes (remifentanyl remifentanil)
• MeSH auto-expansion: Automatic conversion to Medical Subject Headings + synonyms
• Advanced search filters: All PubMed field tags that human users can set are exposed to agents (age groups, sex, species, language, publication types, clinical queries, etc.)
• PICO parsing: Clinical question decomposition (Patient/Intervention/Comparison/Outcome)
• Multi-source parallel search: Simultaneous queries across PubMed, Europe PMC, CORE, Semantic Scholar, OpenAlex
• Agent-friendly output: Structured responses optimized for LLM decision-making

The goal is to make literature research as accessible to AI agents as it is to human researchers.

Feedback welcome!
I'm happy to make adjustments based on your suggestions. If there are specific modifications needed for better TU integration, please let me know!

[gasvn]

Thank you, @u9401066! These are great tools. Given that the dependencies are lightweight and we already have some PubMed-related tools, would it be possible to make them local tools in ToolUniverse? You can refer to this doc for guidance: https://github.com/mims-harvard/ToolUniverse/blob/main/tool_implementation_guide.md
Since remote tools need to be activated by users or require the MCP server to already be available, it would be better to make them local tools to ensure broader usage.