Skip to main content
Glama
med-lit-mcp

med-lit-mcp

by fc0web

med-lit-mcp

An MCP server to cut down the time clinicians spend "looking things up." It lets you search PubMed, clinical trials, drug information, clinical practice guidelines, and the medical fee schedule master directly from Claude.

A U.S. survey found that physicians spend 1.5–2 hours on administrative tasks and paperwork for every hour of patient care. Part of that time goes to checking the literature, package inserts, and regulations. This server targets exactly that.


What this server does not do

This software is not a medical device software (SaMD). Under the PMD Act, regulatory applicability is determined by "the degree of contribution to medical judgment" and "risk (harm in the event of a malfunction)." This server only points to the location of primary information and makes no medical judgment, so it stays within scope of non-applicability.

By design, it does not:

  • Present a diagnosis or list differential diagnoses

  • Recommend treatment plans, drug choices, or dosages

  • Make interpretations based on lab results or images

  • Summarize/translate abstracts or package inserts, extends the original text

All it returns is the source and the original text. Medical judgment is the responsibility of the clinician.

This policy is enforced in code. src/med_lit_mcp/guard.py contains patterns of prohibited language, and tests verify that server-generated text (tool descriptions, disclaimers, error messages) does not contain expressions that suggest a medical decision.

Related MCP server: Medical MCP Server

What data it handles

It does not handle any patient personal information. What is sent externally is only the search keyword. Communication is restricted to the four hosts below; any other destination raises an exception (see ALLOWED_HOSTS in http.py).

Host

Use

eutils.ncbi.nlm.nih.gov

PubMed (NCBI E-utilities)

clinicaltrials.gov

ClinicalTrials.gov API v2

rest.kegg.jp

KEGG DRUG (drug information)

shinryohoshu.mhlw.go.jp

Medical fee schedule information service (master)

Retrieved results are cached in local SQLite (default ~/.cache/med-lit-mcp/). Once a medical fee schedule master is imported, those searches work without a network.

Installation

uvx --from git+https://github.com/fc0web/med-lit-mcp med-lit-mcp

Alternatively, clone it:

git clone https://github.com/fc0web/med-lit-mcp
cd med-lit-mcp
uv sync
uv run med-lit-mcp

Registration with Claude Desktop / Claude Code

Add it manually to either claude_desktop_config.json (Claude Desktop) or .mcp.json (Claude Code):

{
  "mcpServers": {
    "med-lit": {
      "command": "uvx",
      "args": ["--from", "git+https://github.com/fc0web/med-lit-mcp", "med-lit-mcp"],
      "env": {
        "NCBI_EMAIL": "you@example.com"
      }
    }
  }
}

For Claude Code, you can also register it with:

claude mcp add med-lit -- uvx --from git+https://github.com/fc0web/med-lit-mcp med-lit-mcp

Tools

Tool

Description

search_pubmed

PubMed search. Filter by author, journal, year, and study design (RCT / meta-analysis / systematic review, etc.)

fetch_pubmed_abstract

Get the full abstract by PMID. Keep the sections of a structured abstract.

search_clinical_trials

ClinicalTrials.gov search. Filter by condition, intervention, availability, and country

find_guideline

Search clinical practice guidelines. Filters with PubMed guideline[pt] and provides a Minds search URL.

search_drug

Drug search by generic or brand name (includes KEGG DRUG and Japanese Pharmacopoeia-listed drugs)

fetch_drug_entry

Drug details (classification, target, metabolism, interactions)

import_shinryo_master

Import a medical fee schedule master (first time only)

search_shinryo_hoshu

Search point values and drug prices (works offline)

shinryo_master_status

Check the imported master and coverage date

Usage examples

「オルフォルグリプロンの第3相試験を、2025年以降のRCTに絞って探して」
  → search_pubmed(term="orforglipron", year_from=2025, study_types=["rct"])

「初診料の点数を調べて」
  → import_shinryo_master(kind="shinryo")   # 初回のみ
  → search_shinryo_hoshu(term="初診料")

Settings (environment variables)

Variable

Default

Description

NCBI_API_KEY

None

When set, NCBI rate limit increases from 3 to 10 req/s

NCBI_EMAIL

None

Contact email that NCBI recommends providing. Recommended.

MED_LIT_CACHE

1

Set to 0 to disable caching

MED_LIT_TIMEOUT

30

HTTP timeout (seconds)

XDG_CACHE_HOME

~/.cache

cache directory

No API key is needed. When the rate limit is exceeded, the server waits and retries rather than throwing an exception.

When using a corporate proxy or a self-signed CA

MCP clients launch the server as a subprocess and only pass a minimal set of environment variables. If your corporate network relies on a private CA, TLS verification may fail. Be sure to set them explicitly in the configuration's env:

"env": {
  "SSL_CERT_FILE": "/path/to/ca-bundle.crt",
  "REQUESTS_CA_BUNDLE": "/path/to/ca-bundle.crt",
  "HTTPS_PROXY": "http://proxy.example.jp:8080"
}

What it can't do (honestly)

  • Full-text search of the Minds guideline library. Minds renders search results in the browser and has no public API. It is technically possible to fetch it with a headless browser, but it's fragile under specification changes and terms of use are not clear, so v0.1 only provides the Minds search URL.

  • Full-text body of PMDA package inserts. There is likewise no public API, so KEGG is the primary source and the PMDA search URL is made available. For clinical use, always refer to the PMDA's official package inserts.

  • Master data for dentistry and dispensing. v0.1 covers three categories only: medical procedures, drug products, and specific medical devices.

Development

uv sync
uv run pytest              # 外部APIを叩かないテスト
uv run pytest -m live      # 実APIへの疎通テスト
uv run ruff check .

Sources

License

MIT

Disclaimer

This software does not provide medical device software. It is not intended for diagnosis or treatment decisions, and the user is solely responsible for final verification of the output. Because regulations, fee schedules, drug prices, and guidelines are subject to change, always confirm with primary sources in practice.

Available Tools

9 tools
fetch_drug_entry医薬品詳細取得A

KEGG IDを指定して医薬品の詳細(一般名・商品名・分類・標的・相互作用など)を取得します。search_drug で得た kegg_id を渡してください。

Inputs: kegg_id (required). Returns: object

fetch_pubmed_abstractPubMed 抄録取得A

PMIDを指定して抄録の全文と書誌情報を取得します。抄録は原文のまま返します(要約・翻訳・解釈は行いません)。構造化抄録はセクションごとに分けて返します。

Inputs: pmids (required). Returns: object

find_guideline診療ガイドライン探索A

あるトピックの診療ガイドラインを探します。PubMedを guideline[pt] で絞った結果と、Mindsガイドラインライブラリの検索URLを返します。Mindsは検索結果をブラウザ側で描画しており機械的に取得できないため、URLの提示に留めています。

Inputs: topic (required), limit, year_from. Returns: object

import_shinryo_master診療報酬マスター取り込みA

診療報酬情報提供サービスから公式マスターを取得してローカルDBに取り込みます。初回のみ実行が必要で、以後の検索はオフラインで動きます。kind: shinryo(医科診療行為)/ iyakuhin(医薬品)/ tokutei(特定器材)。

Inputs: kind, force. Returns: object

search_clinical_trials臨床試験検索B

ClinicalTrials.gov を検索します。疾患・介入・実施状況・相・国で絞り込めます。進行中の試験や、ある薬剤がどの段階にあるかを確認するのに使います。

Inputs: term (required), limit, phase, status, +3 more. Returns: object

search_drug医薬品検索A

一般名・商品名から医薬品を検索します(KEGG DRUG)。日本薬局方の収載状況や日本の商品名も引けます。添付文書の本文はPMDAで確認する必要があるため、検索URLを併記します。

Inputs: term (required), limit. Returns: object

search_pubmedPubMed 文献検索A

PubMedを検索して書誌情報の一覧を返します。キーワード・著者・雑誌・年・研究デザインで絞り込めます。抄録の全文が必要なときは fetch_pubmed_abstract を続けて呼んでください。結果には必ず出典URLと取得日時が付きます。

Inputs: term (required), sort, limit, author, +5 more. Returns: object

search_shinryo_hoshu診療報酬・薬価検索C

取り込み済みの診療報酬マスターを検索します。名称・カナ・コードで引けます。ネットワークを使わないためオフラインでも動作します。点数は改定で変わるため、収載ファイル名を結果に添えています。

Inputs: term (required), kind, limit. Returns: object

shinryo_master_statusマスター取り込み状況A

ローカルに取り込み済みの診療報酬マスターの一覧と収載日を返します。

Inputs: no named parameters. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A3.7/5.0

    Scored across 9 tools

    Disambiguation5/5

    Each tool targets a distinct resource and action: drug entries, PubMed searches, abstract retrieval, clinical trial search, guideline discovery, drug search, and shinryo master operations are all clearly separated. Even the two drug-related tools (search_drug and fetch_drug_entry) have a clear search-then-fetch workflow, avoiding overlap.

    Naming Consistency4/5

    The naming follows a mostly consistent pattern with verb prefixes (search_, fetch_, import_, find_), though shinryo_master_status deviates by being a noun phrase instead of an action. Minor inconsistency with 'find_guideline' versus 'search_' but overall the pattern is predictable and readable.

    Tool Count5/5

    With 9 tools, the server is well-scoped for its purpose of medical literature and drug/reimbursement information retrieval. Each tool covers a distinct domain or step in a workflow, and the count feels neither bloated nor insufficient for the stated functionality.

    Completeness4/5

    The surface covers the main workflows: drug search/detail, PubMed search/abstract, clinical trial lookup, guideline discovery, and shinryo master management. Minor gaps exist, such as lacking full-text article retrieval beyond abstracts or direct PMDA label fetching, but these are acknowledged as external limitations and agents can work around them.

    Maintenance

    ActivityStale
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI assistants to query authoritative medical sources such as FDA, WHO, PubMed, RxNorm, and ClinicalTrials.gov for drug information, safety data, health statistics, clinical guidelines, pediatric resources, and evidence-graded literature. Runs locally with no API keys or data leaving the machine.
      83 npm
      ISC
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI clients to query authoritative medical sources such as FDA, WHO, PubMed, RxNorm, and ClinicalTrials.gov for drug information, medical literature, health statistics, clinical guidelines, and pediatric resources through local MCP tools.
      83 npm
      ISC
    • A
      license
      Not graded
      quality
      A
      maintenance
      Enables AI assistants to query authoritative sources like FDA, WHO, PubMed, RxNorm, and Google Scholar for drug information, health statistics, clinical guidelines, trials, and medical literature. It runs locally with no API keys, so data stays private while results come back with evidence grading and resilient fallbacks.
      83 npm
      ISC