Provides tools for searching PubMed literature, retrieving article details, extracting key information, finding related articles, managing cache, and downloading open-access full-text PDFs via the PubMed, PMC, and Unpaywall APIs.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@PubMed MCP Server search for recent articles on CRISPR"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
为 LLM Agent 提供结构化 PubMed 文献数据的 MCP 服务器。Agent 友好的响应模型，专注数据提供，分析交给 LLM。
LLM Agent <--MCP--> PubMed MCP Server <--API--> PubMed / PMC / Unpaywall核心能力： 文献搜索 / 智能缓存 / OA 全文下载 / Agentic 响应模型
前置要求： Node.js v18.0.0+
# npm 全局安装
npm install -g mcp-pubmed-llm-server
# 或从源码构建
git clone git@github.com:PancrePal-xiaoyibao/mcp-pubmed-server-pancrpal.git
cd mcp-pubmed-server-pancrpal
npm install && npm run buildcp .env.example .env编辑 .env：
PUBMED_API_KEY=你的NCBI_API密钥 # 可选，https://www.ncbi.nlm.nih.gov/account/settings/
PUBMED_EMAIL=你的邮箱 # 可选（建议填写）
ABSTRACT_MODE=quick # quick(1500字符) | deep(6000字符)
FULLTEXT_MODE=disabled # disabled | enabled | auto说明： API Key 和 Email 均非必填。无 Key 时匿名运行（3 次/秒），有 Key 时 10 次/秒。
# npm 包
mcp-pubmed-llm-server
# 或
npx mcp-pubmed-llm-server
# 源码开发
npm run dev
# 源码生产
npm run build && npm startRelated MCP server: mcp-pubmed-server
模式 | 适用场景 | 启动方式 |
stdio | 本地 MCP 客户端集成 |
|
Streamable HTTP | 服务端远程部署 |
|
npm start
# 或
node dist/index.js --mode=stdionpm run start:http
# 或
node dist/index.js --mode=streamableHttpDocker 部署：
cd docker
cp .env.example .env # 编辑填入配置
docker compose up -d --build验证：
curl http://localhost:8745/health
# {"status":"ok","mode":"streamableHttp","sessions":0}端点：
POST /mcp — MCP 协议消息
GET /mcp — SSE 事件流
DELETE /mcp — 关闭会话
GET /health — 健康检查
{
"mcpServers": {
"pubmed": {
"command": "npx",
"args": ["-y", "mcp-pubmed-llm-server"],
"env": {
"PUBMED_API_KEY": "你的API密钥（可选）",
"PUBMED_EMAIL": "你的邮箱（可选）",
"ABSTRACT_MODE": "deep",
"FULLTEXT_MODE": "enabled"
}
}
}
}配置文件位置：
Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
Claude Code: ~/.claude/config.json
Cline: VS Code 设置中的 MCP Servers
stdio 模式： 同上配置，type 设为 stdio
streamableHttp 模式： type 设为 streamableHttp，baseUrl 设为 http://<服务器IP>:8745/mcp
工具 | 说明 | 关键参数 |
| 文献搜索，支持 Boolean/MeSH |
|
| 获取 PMID 完整信息（单个或批量） |
|
| 提取论文关键信息段落 |
|
| 查找相关/综述文献 |
|
工具 | 说明 | 关键参数 |
| 缓存统计、清理、清空 |
|
FULLTEXT_MODE=enabled）工具 | 说明 | 关键参数 |
| 检测 OA 状态和全文可用性 |
|
| 下载全文 PDF（单篇或批量） |
|
| 系统环境和 API Key 状态检测 | 无 |
每个工具返回统一的 AgentResponse<T> 结构，为 AI Agent 优化：
{
"status": "success",
"data": { ... },
"metadata": {
"tool": "pubmed_search",
"executionMs": 1234,
"timestamp": "2025-01-01T00:00:00.000Z",
"pagination": { "total": 500, "returned": 20, "hasMore": true }
},
"suggestions": [
{
"tool": "pubmed_get_details",
"reason": "Get full metadata for specific articles of interest.",
"parameters": { "pmids": ["12345678"] }
}
]
}status — 成功/错误状态
data — 类型化的返回数据
metadata — 执行上下文（耗时、分页、缓存状态）
suggestions — Agent 下一步操作建议（工具名 + 原因 + 参数）
支持多个 NCBI API Key 轮询/主备/随机负载均衡。
在项目根目录创建 api-keys.json（参见 api-keys.json.example）：
{
"keys": [
{ "api_key": "KEY_1", "email": "user1@example.com" },
{ "api_key": "KEY_2", "email": "user2@example.com" }
],
"strategy": "round-robin"
}策略 | 说明 |
| 轮询（默认） |
| 主备切换 |
| 随机负载均衡 |
优先级：
api-keys.json> 环境变量 > 匿名模式健康管理： 连续 3 次失败自动下线，60 秒冷却后恢复
src/
├── index.ts # 入口点
├── config.ts # 配置常量 + 环境变量
├── server.ts # MCP Server 编排器
├── types/ # TypeScript 类型定义
│ ├── article.ts # Article, SearchResult, OAInfo
│ ├── responses.ts # AgentResponse<T>, makeResponse/makeError
│ └── index.ts
├── api/
│ ├── pubmed-client.ts # PubMed EUtilities 客户端
│ └── key-pool.ts # API Key 号池（轮询/主备/随机）
├── cache/
│ ├── memory-cache.ts # 内存 LRU 缓存（5 分钟，100 条上限）
│ └── file-cache.ts # 文件持久化缓存（30 天过期）
├── services/
│ ├── fulltext.ts # OA 检测 + PDF 下载
│ └── system.ts # 系统环境检测
├── tools/
│ ├── definitions.ts # MCP 工具 Schema（8 个工具）
│ └── handlers.ts # 工具路由 + 处理逻辑
├── transport/
│ ├── stdio.ts # stdio 传输
│ └── streamable-http.ts # HTTP 传输（Express）
└── utils/
└── formatter.ts # 文章格式化（compact/standard/detailed）问题 | 解决方案 |
依赖缺失 |
|
PubMed API 调用失败 | 检查网络；有 Key 时确认 Key 有效 |
Key 池全部不可用 | 检查 |
端口被占用 |
|
Docker 健康检查失败 | 检查 |
MIT License
Extract specific structured sections from a PubMed article. Use when you need only certain aspects (author details, structured abstract, keywords, DOI) rather than the full record. More token-efficient than get_details for targeted extraction.
| Name | Required | Description | Default |
|---|---|---|---|
| pmid | Yes | PubMed article ID. | |
| sections | No | Which sections to extract. |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, openWorldHint, and idempotentHint, covering safety and side-effect expectations. The description adds context about token efficiency and the targeted nature of extraction, but does not disclose return format or further behavioral details. This is adequate but not rich beyond annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences, front-loaded with purpose, then usage guidance. No filler or redundancy. Every sentence contributes meaning.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the moderate complexity (2 params, no output schema, read-only), the description provides enough context: what it extracts, when to use, and how it compares to a sibling. It does not explain return values, but the annotations cover safety and the description conveys the core use case well.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%: both 'pmid' and 'sections' have descriptive text in the schema. The description mentions examples of sections (authors, abstract, keywords, DOI) which echo the schema enums, adding no new meaning. Baseline 3 is appropriate since the schema handles parameter semantics fully.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool's function: 'Extract specific structured sections from a PubMed article.' It uses a specific verb ('extract') and resource ('sections from a PubMed article'), and differentiates from siblings by noting it is for partial data rather than the full record, explicitly contrasting with get_details.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides explicit guidance on when to use: 'Use when you need only certain aspects… rather than the full record.' It also names the alternative (get_details) and highlights token efficiency. It does not explicitly state when not to use, but the 'when' scenario is clear enough.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Retrieve complete metadata for specific PubMed articles by PMID. Accepts a single PMID or array of PMIDs (up to 20). Returns full article records including abstract, authors, journal, DOI, MeSH terms. Use when you have specific PMIDs from a previous search or citation and need the full record.
| Name | Required | Description | Default |
|---|---|---|---|
| pmids | Yes | Single PMID string or array of PMIDs. | |
| format | No | Output format. | standard |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare read-only, open-world, and idempotent behavior. The description adds the limit of up to 20 PMIDs and the specific fields returned (abstract, authors, journal, DOI, MeSH terms), which is useful context beyond the annotations. It doesn't mention error handling, but the annotation coverage lowers the burden.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is three sentences, front-loaded with the main action, and contains no filler or redundant information. Every sentence provides value.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's moderate complexity, with 2 parameters and annotations but no output schema, the description covers the input format, limits, expected return content, and usage context. This is sufficient for an agent to invoke it correctly.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% with both parameters described. The description only restates what the schema already says (single or array, up to 20), adding no additional parameter-specific meaning. Baseline 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states 'Retrieve complete metadata for specific PubMed articles by PMID', which uses a specific verb and resource. It distinguishes from siblings by focusing on PMID-based retrieval rather than searching, and mentions 'complete metadata' which is distinct from other tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
It says 'Use when you have specific PMIDs from a previous search or citation and need the full record', providing clear context. However, it doesn't explicitly name alternative tools (e.g., pubmed_search) or state when not to use it, so it falls short of full explicit guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
View cache statistics or perform maintenance. Use action="stats" to check hit rates and storage usage. Use action="clean" to remove expired entries, or action="clear" to wipe a specific cache layer.
| Name | Required | Description | Default |
|---|---|---|---|
| action | No | Cache operation to perform. | stats |
| target | No | Which cache layer to operate on. | all |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate destructiveHint=true. The description adds valuable nuance by showing that 'stats' is safe, 'clean' removes expired entries (moderate), and 'clear' wipes a cache layer (destructive). It clarifies the specific destructive behavior for each action, which is more granular than the single annotation flag.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is two sentences, front-loaded with the core purpose, followed by concise action-specific guidance. Every word earns its place – no filler or redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the tool's simplicity (2 enum params, no output schema, annotations present), the description covers the key behavioral aspects: the three actions, their intent, and the existence of cache layers via 'target'. It could mention that 'clear' is irreversible or what 'stats' returns, but the current description is sufficient for basic correct use.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% with enum parameters already explaining each field. The description adds semantics by mapping actions to real-world outcomes (checking hit rates, removing expired entries, wiping layers), which deepens the meaning beyond the generic 'Cache operation to perform.' This extra context helps the agent choose the right action without additional lookups.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description starts with 'View cache statistics or perform maintenance,' clearly stating the tool's purpose as cache management. It explicitly distinguishes itself from sibling tools (search, details, extract, related) by focusing on internal DB cache operations. The actions are enumerated with specific outcomes (hit rates, storage usage, expired entries, wipe layer).
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides clear usage context by explaining when to use each action: 'stats' for hit rates and storage, 'clean' for removing expired entries, 'clear' for wiping a layer. While it doesn't explicitly mention alternatives to this tool, the sibling tools are obviously unrelated to cache management, making the intended use case clear.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search PubMed biomedical literature database. Returns articles matching your query with metadata, abstracts, and citations. Supports Boolean operators (AND, OR, NOT), MeSH terms, and field tags like [Title], [Author]. Use format="compact" for quick overviews when browsing, "detailed" for in-depth analysis. Use days_back to limit to recent publications.
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | PubMed search query. Supports Boolean logic and MeSH terms. | |
| format | No | Response detail level. compact=minimal, standard=balanced, detailed=full metadata+structured abstract. | standard |
| sort_by | No | Sort order for results. | relevance |
| days_back | No | Only return articles published within the last N days. 0 = no limit. | |
| max_results | No | Number of results to return (1-100). |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare readOnlyHint, idempotentHint, and openWorldHint, so safety is covered. The description adds value by disclosing that results include metadata, abstracts, and citations, and by mentioning supported query features like Boolean operators and MeSH terms. No contradictions with annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is four sentences, front-loaded with the core purpose, and every sentence provides useful information without redundancy. It is compact and well-structured.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the main output contents and several key parameters, and the schema covers the remaining parameters. With no output schema, the description explains that results contain metadata, abstracts, and citations. It could arguably mention sort_by or max_results usage, but it is otherwise complete for a search tool.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, providing a baseline of 3. The description adds meaningful extra detail beyond the schema for query, mentioning field tags like [Title] and [Author], and gives usage guidance for format and days_back, which helps the agent choose parameter values.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states 'Search PubMed biomedical literature database' and describes the output as 'articles matching your query with metadata, abstracts, and citations.' This specific verb+resource phrasing distinguishes it from sibling tools like pubmed_get_details or pubmed_find_related.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides clear in-tool guidance for format and days_back, but it does not explicitly say when to use this tool versus alternatives like pubmed_get_details or pubmed_find_related. Usage guidance is mostly implied by the search-oriented purpose.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Each tool has a clear primary purpose: search, fetch full records, extract specific fields, find related articles, and manage cache. However, pubmed_get_details and pubmed_extract_info could be confused by an agent, as both retrieve article information, though the descriptions differentiate full records from targeted extraction.
All tool names follow a consistent pattern: 'pubmed_' prefix followed by a verb or verb_noun phrase (search, get_details, extract_info, find_related, manage_cache). The naming is uniform, snake_case, and clearly indicates the operation performed.
Five tools is a well-scoped set for a PubMed MCP server, covering core literature search and retrieval without redundancy. The count feels appropriately moderate, not overwhelming or thin.
The toolset covers the full typical workflow: search for articles, retrieve full details, extract specific sections, discover related articles, and manage cache for efficiency. No obvious critical gaps exist for a literature search domain.
Unclaimed servers have limited discoverability.
If you are the server author, to access and configure the admin panel.
Search PubMed/Europe PMC, fetch articles and full text (PMC/EPMC/Unpaywall), citations, MeSH terms.
PubMed MCP — wraps the NCBI E-utilities API (biomedical literature, free, no auth)
Search biomedical literature, get article details, find related articles, and explore MeSH terms
Search biomedical papers, inspect publication records, and traverse citation or semantic graphs.
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/PancrePal-xiaoyibao/mcp-pubmed-server-pancrpal'
If you have feedback or need assistance with the MCP directory API, please join our Discord server