feat(connections): add connections export, contact enrichment, company search, notifications

[Gabrcodes]

Summary

Four new tools for connections and notifications, plus a network degree filter for search_people.

Tool	Description
get_my_connections	Scrapes the connections list via infinite scroll. Uses innerText parsing only — no CSS class-name selectors, compliant with project scraping rules.
extract_contact_details	Batch-enriches profiles with email, phone, website, birthday from the contact info overlay. Chunks with configurable delays. include_raw=false by default to avoid large responses.
get_connections_at_company	Resolves the company name to a numeric entity ID first (LinkedIn ignores name strings in currentCompany filter), then searches 1st-degree connections. Falls back to keyword search if ID can't be resolved. Closes #248.
get_notifications	Scrapes /notifications/ for all notification types. Closes #211.

search_people gains a network parameter ("F" = 1st degree, "S" = 2nd, "O" = 3rd+) with input validation.

Implementation notes

• _resolve_company_id navigates to the company page and extracts the numeric entity URN from page metadata — needed because currentCompany=%5B%22Google%22%5D is silently ignored by LinkedIn
• scrape_contact_batch catches RateLimitError specifically to abort the batch; other LinkedInScraperException failures are per-profile only
• _parse_contact_record extracts company from the headline ("X at Company") rather than guessing from innerText line positions

Changes

• scraping/extractor.py — _parse_contact_record, search_people network filter, scrape_connections_list, scrape_contact_batch, get_notifications, _resolve_company_id, search_connections_at_company
• tools/connections.py — new file with 4 tool registrations
• tools/person.py — network parameter on search_people
• server.py — register connections tools
• tests/test_tools.py — updated search_people assertion for network param

Test plan

• 357 passed, 5 skipped, 0 failures
• ruff check and ruff format pass
• Live test: get_my_connections returns connection list
• Live test: get_connections_at_company with a real company
• Live test: get_notifications

[greptile-apps]

Greptile Summary

This PR adds four new tools (get_my_connections, extract_contact_details, get_connections_at_company, get_notifications) and a network degree filter for search_people. The implementation is well-structured, follows the project's innerText-first scraping rules, handles RateLimitError correctly in the batch flow, and the company-ID resolution approach is sound. All previously flagged concerns (quote_plus → quote, double-navigation timeout) appear to have been addressed.

Confidence Score: 5/5

Safe to merge; all remaining findings are minor P2 style suggestions with no correctness impact.

No P0 or P1 issues found. The two open comments are best-practice suggestions (chunk_delay guard and notifications limit semantics), neither of which affects correctness or reliability in normal use. Prior concerns about quote_plus and timeout have been resolved.

extractor.py — chunk_delay guard and notification limit approximation are minor but worth a follow-up.

Important Files Changed

Filename	Overview
linkedin_mcp_server/scraping/extractor.py	Adds scrape_connections_list, scrape_contact_batch, get_notifications, _resolve_company_id, and search_connections_at_company; network filter on search_people. Minor: chunk_delay has no lower bound; notification limit controls scroll depth but doesn't clip the returned text.
linkedin_mcp_server/tools/connections.py	New file registering four MCP tools; error handling is consistent with existing tools; timeout values look appropriate with the * 2 multiplier for get_connections_at_company and * 5 for extract_contact_details.
linkedin_mcp_server/tools/person.py	Adds network parameter to search_people tool; validation is delegated to the extractor layer which raises ValueError for invalid values.
linkedin_mcp_server/server.py	Minimal change: imports and registers the new connections tools module.
tests/test_tools.py	Updates search_people assertion to account for the new network parameter; no regressions visible.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[get_my_connections] --> B[scrape_connections_list]
    B --> C[Navigate connections page]
    C --> D[scroll_to_bottom]
    D --> E[evaluate: extract /in/ links]
    E --> F[Return connections list]

    G[extract_contact_details] --> H[scrape_contact_batch]
    H --> I{For each chunk}
    I --> J[extract_page profile]
    J --> K[_extract_overlay contact-info]
    K --> L[_parse_contact_record]
    L --> M{More chunks?}
    M -->|Yes, sleep chunk_delay| I
    M -->|No| N[Return contacts + failed]

    O[get_connections_at_company] --> P[search_connections_at_company]
    P --> Q[_resolve_company_id]
    Q --> R{ID found?}
    R -->|Yes| S[Search with network + currentCompany filter]
    R -->|No| T[Fallback keyword search]

    U[get_notifications] --> V[Navigate notifications page]
    V --> W[scroll_to_bottom]
    W --> X[_extract_root_content main]
    X --> Y[Return notifications text]

    Z[search_people] --> AA{network param?}
    AA -->|Valid F S O| AB[Add network filter to URL]
    AA -->|Invalid| AC[Raise ValueError]
    AB --> AD[extract_page search results]

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: linkedin_mcp_server/scraping/extractor.py
Line: 288-295

Comment:
**`chunk_delay` has no lower-bound guard**

`asyncio.sleep(chunk_delay)` is called unconditionally with a caller-supplied float. There's no validation in either `scrape_contact_batch` or the tool layer, so a caller (or an LLM) can pass `chunk_delay=0` and silently remove the inter-chunk pause that protects against rate limiting.

```suggestion
            if chunk_idx + chunk_size < total and chunk_delay > 0:
```

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: linkedin_mcp_server/scraping/extractor.py
Line: 320-322

Comment:
**`limit` controls scroll depth, not result count**

`scrolls = max(1, limit // 10)` is an approximation — the final raw text returned is not clipped to `limit` notifications. With `limit=1`, `scrolls=1` but the page may render far more than 1 notification; with `limit=50`, up to 50 may be loaded but the caller gets back the full `<main>` text with no numeric cutoff. The parameter name and description ("Maximum notifications to load") imply a stricter bound than what's implemented. Consider documenting this in the docstring, or renaming the parameter to `max_scrolls` to match `scrape_connections_list`.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (3): Last reviewed commit: "fix: propagate section_errors and omit e..." | Re-trigger Greptile}