Academic Literature MCP (Multi-site)
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| ACADEMIC_MCP_WOS_DB | No | Web of Science database abbreviation. | WOS |
| ACADEMIC_MCP_DATA_DIR | No | Directory for the database, downloads, exports, and browser profile. | ./data |
| ACADEMIC_MCP_WOS_API_KEY | Yes | Clarivate API key required for Web of Science search. | |
| ACADEMIC_MCP_MAX_INFLIGHT | No | Maximum number of concurrent requests. | 3 |
| ACADEMIC_MCP_VIP_BASE_URL | No | VIP base URL. | |
| ACADEMIC_MCP_VIP_PASSWORD | No | VIP password. | |
| ACADEMIC_MCP_VIP_USERNAME | No | VIP username. | |
| ACADEMIC_MCP_WOS_API_BASE | No | Web of Science API base URL, useful for proxies or testing. | official Starter API |
| ACADEMIC_MCP_ENABLED_SITES | No | Comma-separated list of enabled sites. | wos,wanfang,vip |
| ACADEMIC_MCP_VIP_LOGIN_MODE | No | VIP login mode. | |
| ACADEMIC_MCP_BROWSER_CHANNEL | No | Browser channel; auto uses Edge then Chrome, or force msedge/chrome. | auto |
| ACADEMIC_MCP_CREDENTIAL_FILE | No | Credential JSON file in plaintext; 600 permissions are recommended. | data/credentials.json |
| ACADEMIC_MCP_UNPAYWALL_EMAIL | No | Contact email for open-access lookup. | |
| ACADEMIC_MCP_VIP_INSTITUTION | No | VIP institution. | |
| ACADEMIC_MCP_BROWSER_HEADLESS | No | Run the browser headless; headed mode is recommended for login and verification. | false |
| ACADEMIC_MCP_WANFANG_BASE_URL | No | Wanfang base URL. | |
| ACADEMIC_MCP_WANFANG_PASSWORD | No | Wanfang password. | |
| ACADEMIC_MCP_WANFANG_USERNAME | No | Wanfang username. | |
| ACADEMIC_MCP_SESSION_PROBE_TTL | No | Session status cache TTL in seconds. | 120 |
| ACADEMIC_MCP_BROWSER_EXECUTABLE | No | Browser executable path for non-standard installations. | |
| ACADEMIC_MCP_WANFANG_LOGIN_MODE | No | Wanfang login mode. | |
| ACADEMIC_MCP_WOS_ENABLE_BROWSER | No | Enable the institutional-subscription PDF channel for Web of Science. | false |
| ACADEMIC_MCP_INLINE_WAIT_SECONDS | No | Wait time within a tool call, in seconds. | 40 |
| ACADEMIC_MCP_REQUEST_INTERVAL_MS | No | Request interval in milliseconds. | 600 |
| ACADEMIC_MCP_WANFANG_DETAIL_BASE | No | Wanfang detail page base URL. | |
| ACADEMIC_MCP_WANFANG_INSTITUTION | No | Wanfang institution. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| list_sitesA | 列出已注册站点及能力矩阵，用于发现哪些站点已可检索/下载。 |
| searchC | 按研究问题检索学术文献（默认 Web of Science）。 |
| structured_searchB | 布尔表达式精确检索：叶子 {field,value,match}，组合 {op:and|or|not,items}。 字段：title、topic、abstract、author(s)、source、doi、document_type、 issn、accession、organization、pmid、year。match 为 phrase（默认）或 exact。 |
| search_resultsB | 本地分页读取检索快照，不联网。 scope：selected（筛选结果）/ eligible（全部相关）/ candidates（全部候选） view：titles / compact / records |
| get_metadataA | 批量收集题录：按 record_ref、完整题名或 search_id 获取摘要/关键词/DOI 等。 local_only=true 时不联网；refresh=true 忽略 30 天缓存重读详情。 quality=true 时尝试补充期刊指标（WoS Starter API 无 JCR，会明确标注不可得）。 |
| download_paperA | 下载全文（PDF/CAJ）。仅在用户确实需要文件时调用。 auto 优先 PDF，站点只提供 CAJ 时回落 CAJ（知网常见）。 解析链：机构订阅浏览器会话 → 开放获取副本（Unpaywall）→ 出版商页面。 不做付费墙绕过；结果未知时不会自动重放点击，需显式 retry=true。 |
| export_citationsC | 导出参考文献：GB/T 7714、APA、MLA、Chicago、Vancouver、BibTeX、RIS、 EndNote、CSL-JSON、JSON、CSV、Markdown、custom。 |
| session_statusA | 查看站点登录/会话状态；refresh=true 时用浏览器会话重新探测。 |
| loginA | 打开独立浏览器窗口，由用户完成机构/个人登录。 万方/维普必须显式传 site（如 site="wanfang"），否则默认打开 WoS 登录页。 force=true 会清除已保存的会话后重新登录。 |
| operation_statusA | 查询任务进度与结果。 wait_seconds 最多 55 秒；任务完成或进入 awaiting_user 会提前返回。 省略 operation_id 时返回当前活动任务。 |
| operation_controlC | cancel 取消；continue 继续中断/预算耗尽的检索（可追加预算）； resume 告知人工步骤完成；show 重显浏览器窗口；recover 重启浏览器。 |
| diagnosticsA | 本地检查：配置、数据量、浏览器依赖、限速与当前任务；不联网。 |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
| literature_review | 围绕一个主题完成检索、筛选、精读与综述提纲 |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| guide | 工具契约、任务状态与证据边界 |
TDQS
Scored across 12 tools
Search vs structured_search is a clear split (natural language vs boolean DSL), and search_results (local snapshot paging) vs get_metadata (external metadata collection) are distinguished by scope and network use. Borderline pairs remain: operation_status vs operation_control both act on tasks, and diagnostics overlaps both by reporting 'current task' plus config/session-adjacent state, which could cause occasional misselection.
Nearly everything is snake_case with a predictable verb_noun or noun_phrase shape (list_sites, get_metadata, download_paper, export_citations, session_status). Deviations are minor: bare nouns/verbs like search, login, diagnostics, and the noun-first trio search_results / structured_search / search don't sit in an obvious ordering.
12 tools is well within a sane range for a multi-site literature workflow: discovery, querying, result reading, metadata, download, citation export, session handling, and task control. Each tool maps to a distinct stage, so nothing feels padded.
The lifecycle is covered end to end: site discovery, plain and boolean search, paged result review, metadata enrichment, full-text download, citation export, login/session, and long-running task control. Gaps are moderate: no way to register or configure a new site despite list_sites exposing a capability matrix, and no cross-site dedup/merge of duplicate records.