Back to Details

academic-mcp-tooling

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Academic MCP Tooling

学术MCP工具

Use MCP servers as source access infrastructure, not as final scholarly authorities. Selection, setup, query discipline, and documentation belong in one auditable workflow so literature access remains reproducible.
将MCP服务器用作资源访问基础设施，而非最终学术权威。选择、设置、查询规范和文档编写应纳入同一个可审计的工作流，以确保文献访问的可复现性。

Read First

必读内容

  • references/mcp-catalog.md
  • references/external-skill-recommendations.md
  • references/source-ledger.md
  • references/repository-contract.md
If the target repo has 
configs/agent-stack.yaml
, read it before selecting MCP servers. If it has 
docs/agent/capability-profile.md
, treat that as the active capability state.
  • references/mcp-catalog.md
  • references/external-skill-recommendations.md
  • references/source-ledger.md
  • references/repository-contract.md
如果目标仓库包含
configs/agent-stack.yaml
，请在选择MCP服务器前阅读该文件。如果包含
docs/agent/capability-profile.md
，请将其视为当前的能力状态依据。

Workflow

工作流

  1. Identify needed source systems: arXiv, Semantic Scholar, OpenAlex, Crossref, PubMed, Zotero, Overleaf, DBLP metadata, or fallback search.
  2. Prefer API-backed source-specific servers over scraping or vague all-in-one wrappers when accuracy matters.
  3. Choose the smallest set of MCP servers that covers the project need.
  4. If 
    academic-research
    is available, inspect the current project state first: 
    academic-research doctor
    .
  5. Document install commands, env vars, auth scopes, rate limits, risks, and workflows in 
    docs/agent/mcp-setup.md
    .
  6. Update 
    docs/agent/capability-profile.md
    when active capability changes.
  7. Mark each server as generated-config, manual-setup, optional, or fallback.
  8. Smoke-test each configured server with one harmless query.
  9. Save useful results to 
    sources/metadata/
    or 
    sota/literature-matrix.csv
    .
  10. Record external IDs, query dates, and deduplication decisions.
  11. Append setup or query changes to 
    wiki/log.md
    .
  1. 确定所需的资源系统：arXiv、Semantic Scholar、OpenAlex、Crossref、PubMed、Zotero、Overleaf、DBLP元数据，或备选搜索方式。
  2. 当准确性至关重要时，优先选择基于API的特定资源服务器，而非爬虫或模糊的一体化包装器。
  3. 选择满足项目需求的最小MCP服务器集合。
  4. 如果
    academic-research
    可用，请先检查当前项目状态：执行
    academic-research doctor
    命令。
  5. 在
    docs/agent/mcp-setup.md
    中记录安装命令、环境变量、授权范围、速率限制、风险及工作流。
  6. 当可用能力发生变化时，更新
    docs/agent/capability-profile.md
    。
  7. 将每个服务器标记为生成配置、手动设置、可选或备选。
  8. 使用一个无风险查询对每个已配置的服务器进行冒烟测试。
  9. 将有用的结果保存到
    sources/metadata/
    或
    sota/literature-matrix.csv
    。
  10. 记录外部ID、查询日期及去重决策。
  11. 将设置或查询的变更追加到
    wiki/log.md
    。

Selection Rules

选择规则

  • Use local PDFs and Zotero records selected by the researcher as the first evidence layer.
  • Prefer arXiv/OpenAlex/Semantic Scholar/Crossref/PubMed metadata over general web search.
  • Use Google Scholar only as fallback discovery unless the user explicitly asks.
  • For Zotero, prefer local-library tools that can access attachments and collections.
  • For Overleaf, require clear token/project setup and default to read-only flows.
  • Disable Sci-Hub, questionable download, or browser-session features unless the user explicitly accepts the legal and institutional risk.
  • 将研究人员选定的本地PDF和Zotero记录作为第一证据层。
  • 优先选择arXiv/OpenAlex/Semantic Scholar/Crossref/PubMed元数据，而非通用网络搜索。
  • 仅在用户明确要求时，才将Google Scholar用作备选发现工具。
  • 对于Zotero，优先选择可访问附件和收藏集的本地库工具。
  • 对于Overleaf，要求明确的令牌/项目设置，默认使用只读流程。
  • 除非用户明确接受法律和机构风险，否则禁用Sci-Hub、有问题的下载或浏览器会话功能。

Smoke-Test Record

冒烟测试记录

For each configured server, record:
  • server name and install command
  • client config location
  • required environment variables
  • query
  • timestamp
  • returned identifier or result count
  • failure, fallback, and known limitation if unavailable
  • active preset or manual rationale when using 
    configs/agent-stack.yaml
  • whether the client config was generated or requires manual setup
对于每个已配置的服务器，记录：
  • 服务器名称和安装命令
  • 客户端配置位置
  • 必填环境变量
  • 查询语句
  • 时间戳
  • 返回的标识符或结果数量
  • 若不可用，记录失败原因、备选方案及已知限制
  • 使用
    configs/agent-stack.yaml
    时的当前预设或手动配置理由
  • 客户端配置是自动生成还是需要手动设置

Query Discipline

查询规范

  • Deduplicate by DOI, arXiv ID, PMID, title, and author/year.
  • Do not rely on one database for final SOTA coverage unless the scope says so.
  • Treat scraped results as discovery leads, not proof.
  • Update 
    sources/source-ledger.csv
    only when a source becomes evidence.
  • 通过DOI、arXiv ID、PMID、标题及作者/年份进行去重。
  • 除非范围明确说明，否则不要依赖单一数据库来覆盖最新研究成果（SOTA）。
  • 将爬虫结果视为发现线索，而非证据。
  • 仅当某一资源成为证据时，才更新
    sources/source-ledger.csv
    。