shield_lock
JinDaGe
← Back to skills
extension
Category: Development & EngineeringNo API key required

ieee-search-mcp

Use browser MCP to access IEEE Xplore through university library proxy, preserve institutional session, run keyword/advanced/journal search, and optionally post-filter by CCF rank (for example CCF-A) with structured output.

personAuthor: jakexiaohubgithub
open_in_newRepositorydownloadDownload package

ieee-search-mcp

Use this skill when user requires IEEE Xplore search through school library access (proxy/SSO session), not direct public IEEE entry.

Preconditions

  • Browser MCP (for example chrome-devtools or equivalent) is already installed and available.
  • The agent has permission to control a browser.
  • If either condition is missing, stop and explicitly say this workflow cannot be executed in-browser.
  • For relative time filters (for example "近2年"), use local time/date tool first and compute explicit year range before filtering.

School Profiles (pluggable)

  1. Load school profile from assets/school_profiles.example.json.
  2. If school is not provided, ask user first.
  3. Treat these as pluggable fields per school:
    • library_home
    • proxy_ieee_home
    • db_keyword_candidates
    • Optional: known_database_detail_url
    • Optional: known_journals — a map of alias -> {name, punumber, proxy_url} for journals the user accesses frequently
  4. Demo/default profile is wust (Wuhan University of Science and Technology).

Common User Intent Paths

Path A: Login + Keyword search + recent years

  1. Open library_home, verify login state.
  2. If login required, ask user to complete SSO/QR and wait.
  3. Search DB by db_keyword_candidates in order (for WUST try IEL before IEEE if needed).
  4. Enter IEEE from library entry and verify institutional session.
  5. Run keyword search.
  6. Apply explicit year range (for example if current year is 2026, "近2年" means 2025-2026).
  7. Return results using requested format.

Path B: Continue from existing proxy IEEE page

  1. If user provides/opened proxy URL (for example metaersp-2.../ieeexplore/...), continue in the same page/session.
  2. Do not open https://ieeexplore.ieee.org/ directly unless user confirms fallback.
  3. Re-verify Access provided by: or proxy host before searching.

Path C: Advanced search / Journal search

  1. Confirm mode with user (keyword / advanced / journal).
  2. Ask for missing constraints (query fields, journal name, years, document type).
  3. Execute search and return structured results.

Path D: Known Journal / Direct Navigation (fast path)

When user mentions a journal by alias or full name and the profile has a known_journals entry:

  1. Match user input against known_journals keys (case-insensitive) or name fields.
  2. If matched, navigate directly to the cached proxy_url — skip Path A search flow entirely.
  3. Verify institutional session (Access provided by: or proxy host in URL).
  4. If session expired, fall through to Path A to re-authenticate, then return to the cached URL.
  5. After reaching the journal page, proceed with any user-requested action (browse recent, search within, etc.).

Learning new journal shortcuts

When the user navigates to a journal not yet in known_journals:

  1. Record the punumber and constructed proxy URL after successful navigation.
  2. Remind the user: "I've noted this journal as a known shortcut — next time just mention it by name and I'll go directly."
  3. The agent should suggest adding to known_journals in the school profile. If the user confirms, update assets/school_profiles.example.json accordingly.

Failure Branches From Real WUST Flow

  • DB alias mismatch: searching IEEE may fail, retry IEL.
  • Click timeout on DB row: refresh snapshot and retry once.
  • Still cannot click: navigate to known detail URL (known_database_detail_url) if profile has it.
  • Session drift: if jumped to public ieeexplore.ieee.org, immediately restore proxy route via proxy_ieee_home.
  • Page/tool argument error: re-check tool parameter type/naming before retry (for example pageId must be number).

Mandatory user checkpoints

Ask user instead of guessing at these points:

  • School or library profile not confirmed.
  • Login step requires human action (captcha/QR/2FA).
  • Multiple near-identical database entries appear and click target is ambiguous.
  • User requests conflicting constraints (for example "only proxy session" and "jump to public site").
  • Output schema not specified but user asks for "整理成表格/导出".

Optional CCF Post-Filter (with ccf-rank skill)

If user asks for "只要 CCF-A/CCF-B" or similar:

  1. Collect venue/journal from IEEE results.
  2. Use sibling skill script: node ../ccf-rank/scripts/query_ccf_rank.mjs "<venue or journal>" --top 3
  3. Keep only rows matching requested rank (for example A).
  4. In output, include both IEEE fields and CCF fields (type, rank, area).

Output Rules

  • Honor user-specified format first.
  • If user requests a table, output Markdown table with requested columns.
  • If user does not specify columns, default columns: title | year | venue/journal | authors | access_path | ieee_url | ccf_rank(optional).
  • Always include access_path so user can confirm proxy route is used.
  • For time-based filters, include explicit year range in response text.

Minimal run summary

For each run, report these 5 items:

  1. Access path used (library_home -> db alias -> IEEE proxy page).
  2. Institutional access preserved or not.
  3. Query mode + query + filters.
  4. Result list (or table).
  5. Next action options (advanced search / journal search / CCF filter / open full text).