eLib_scraper

# eLibrary MCP MCP server for working with the web interface of [eLIBRARY.RU](https://elibrary.ru) from LLM clients (Cursor and others). Transport: **stdio** (`node dist/index.js`). ## Important: Legal Risks Automated access may contradict the rules of eLIBRARY.RU. Use only within your rights and subscription. The repository authors do not guarantee performance when changing the layout or site policy. ## Confidentiality Passwords and cookies are **not** written to the disk by the server; the session is held in the RAM of the process. Do not pass account data in tool arguments — only through environment variables / host secrets. ## Environment Variables | Variable | Purpose | |----------|----------| | `ELIBRARY_BASE_URL` | Base URL (default `https://elibrary.ru`) | | `ELIBRARY_COOKIE` | Mode C: `Cookie` header string after logging in the browser | | `ELIBRARY_LOGIN` / `ELIBRARY_PASSWORD` | Mode A: login and password | | *(without cookie and login)* | Anonymous session: the server requests the main page and receives eLIBRARY cookie (limited access) | | `ELIBRARY_LOGIN_PAGE` | URL or path of the login form page (required for Mode A) | | `ELIBRARY_LOGIN_FIELD` | Login field name if auto-detection does not work | | `MCP_ELIBRARY_TENANT` | Stable user ID from the host (jar isolation in one process) | | `ELIBRARY_GLOBAL_RPS` / `ELIBRARY_TENANT_RPS` | Request per second limits (see `src/config.ts`) | Default entity paths: `item.asp`, `author_profile.asp`, `org_profile.asp`, `journal_profile.asp`. Search: loading `querybox.asp` and POST to `query_results.asp` (see `src/search/`). If the browser has different URLs, check and, if necessary, improve `src/urls.ts`. ## Build and Run ```bash npm ci npm test npm run build npm start ``` Connection in Cursor: `mcpServers` → command `node`, args `["/absolute/path/to/dist/index.js"]`, env with secrets. ## Tools - `elibrary_search` — search (`query`, optionally `page`) - `elibrary_publication`, `elibrary_author`, `elibrary_organization`, `elibrary_journal` — cards by `id` - `elibrary_publication_metrics` — heuristics by page text (citation/metrics) - `elibrary_fulltext` — link to PDF or refusal code - `elibrary_logout` — reset cookie in memory - `elibrary_metrics_snapshot` — success/failure counters by tools All main tools optionally have `tenant_key` if there is no `MCP_ELIBRARY_TENANT`. ## Publication in Smithery / Cloud The current format of `smithery.yaml` is described in the [Smithery documentation](https://smithery.ai/docs). For Playwright-fallback (plan, Task 11), a **container** runtime is usually required. ## MVP Limitations No CAPTCHA bypass; parsers depend on HTML. With `SCHEMA_BREAK`, updating selectors in `src/parsers/` may be required.