Skip to main content

LinkedInIntegration

What it is​

  • A Python integration client for LinkedIn’s private “voyager” API endpoints.
  • Authenticates using LinkedIn cookies (li_at, JSESSIONID) and fetches:
    • Organization info
    • Profile “top card”, skills, experience, education
    • Profile posts feed and post engagement (stats, reactions, comments, reposts)
    • People search and mutual connections exports (JSON + Excel)
  • Persists raw and derived artifacts (JSON, optional images, Excel) under a configured datastore path.
  • Provides as_tools() to expose selected actions as LangChain StructuredTools.

Public API​

LinkedInIntegrationConfiguration (dataclass)​

Configuration for LinkedInIntegration.

  • li_at: str — LinkedIn li_at cookie.
  • JSESSIONID: str — LinkedIn JSESSIONID cookie (quotes are stripped in __init__).
  • linkedin_url: str — a LinkedIn profile URL used to initialize the integration and derive profile_public_id.
  • naas_integration_config: NaasIntegrationConfiguration | None — optional; if provided, enables uploading exported Excel as a Naas asset.
  • base_url: str = "https://www.linkedin.com/voyager/api" — API base.
  • datastore_path: str — storage root (defaults from ABIModule configuration).

LinkedInIntegration​

Main integration client.

Constructor​

  • LinkedInIntegration(configuration: LinkedInIntegrationConfiguration)
    • Validates access by calling get_profile_public_id(configuration.linkedin_url).
    • Raises Exception if the public ID cannot be resolved.

Data retrieval methods​

  • get_organization_id_from_url(url: str) -> Dict[str, str]

    • Parses organization “public id” from /company/, /school/, or /showcase/ URLs.

  • get_organization_id(url: str) -> Dict[str, str]

    • Fetches organization info then extracts the numeric org id from *elements.

  • get_organization_info(url: str, return_cleaned_json: bool = False) -> Dict

    • Calls /organization/companies?... and saves response; optionally returns cleaned/flattened JSON.

  • get_profile_id_from_url(url: str) -> Dict[str, str]

    • Extracts vanity id from /in/<id>/ profile URLs.

  • get_profile_public_id(url: str) -> Dict[str, str]

    • Reads publicIdentifier from top-card response; falls back to parsing a “share URL” from overflow actions.

  • get_profile_id(url: str) -> Dict[str, str]

    • Extracts internal profile id (URN suffix) from top-card response.

  • get_profile_top_card(url: str, return_cleaned_json: bool = False) -> Dict

    • GraphQL request for top-card; rejects vanity ids starting with AcoAA (treated as invalid URL input).

  • get_profile_data(url: str, profile_type: str = "skills", locale: str = "en_US", return_cleaned_json: bool = False) -> Dict

    • GraphQL request for a profile section (skills, experience, education, etc.).

  • get_profile_skills(url: str, return_cleaned_json: bool = False) -> Dict

  • get_profile_experience(url: str, return_cleaned_json: bool = False) -> Dict

  • get_profile_education(url: str, return_cleaned_json: bool = False) -> Dict

  • get_profile_posts_feed(url: str, start: int = 0, count: int = 1, pagination_token: str | None = None, return_cleaned_json: bool = False) -> Dict

    • GraphQL request for profile posts feed.
    • Extracts activity_id and paginationToken; derives a publish date from token.
    • Saves response under a date/activity-based prefix.

  • get_activity_id_from_url(url: str) -> Dict

    • Extracts activity id from URLs containing -activity- or :activity:.

  • get_post_stats(url: str, return_cleaned_json: bool = False) -> Dict

    • Calls /feed/updates/urn:li:activity:{id}.

  • get_post_reactions(url: str, start: int = 0, count: int = 100, limit: int = -1, return_cleaned_json: bool = False) -> Dict

    • GraphQL pagination loop; merges included and data.*elements; saves aggregated JSON.

  • get_post_comments(url: str, start: int = 0, count: int = 100, limit: int = -1, return_cleaned_json: bool = False) -> Dict

    • GraphQL pagination loop; merges comment elements; saves aggregated JSON.

  • get_post_reposts(url: str, start: int = 0, count: int = 100, limit: int = -1, return_cleaned_json: bool = False) -> Dict

    • GraphQL pagination loop; merges reshare elements; saves aggregated JSON.

Search/export methods​

  • get_mutual_connexions(profile_url: str, connection_distance: str = "F", organization_url: str | None = None, start: int = 0, count: int = 50, limit: int = 1000, query_id: str = "...") -> Dict

    • GraphQL pagination loop to fetch people results connected to profile_url (optionally filtered by organization).
    • Builds a simplified list of people (id, public_id, name, headline, location, profile_url, etc.).
    • Saves final JSON and exports Excel; if Naas is configured, uploads and returns an excel_url.

  • search_people(connection_distance: str = "F", organization_url: str | None = None, location: str | None = None, start: int = 0, count: int = 50, limit: int = 1000, query_id: str = "...") -> Dict

    • GraphQL pagination loop for generic PEOPLE search.
    • Optional location mapping currently contains only "France" -> "105015875".
    • Saves final JSON and exports Excel (same upload behavior as above).

Cleaning utilities​

  • clean_json(prefix: str, filename: str, data: dict) -> Dict[str, Any]
    • If a cleaned file exists, returns it.
    • Otherwise:
      • Removes keys starting with * or containing urn (case-insensitive) recursively.
      • Parses included entities into a $type-keyed structure, optionally replacing image-like fields with highest-quality URL.
      • Flattens nested dict keys with _.
      • Saves *_cleaned.json under datastore_path/prefix/.

as_tools(configuration: LinkedInIntegrationConfiguration) -> list​

  • Returns a list of LangChain StructuredTool wrappers around:
    • linkedin_get_organization_info
    • linkedin_get_profile_top_card
    • linkedin_get_profile_skills
    • linkedin_get_profile_experience
    • linkedin_get_profile_education
    • linkedin_get_profile_posts_feed
    • linkedin_get_post_comments
    • linkedin_get_post_reactions
    • linkedin_get_post_reposts
    • linkedin_get_mutual_connexions
    • linkedin_search_people
  • Each tool validates inputs via Pydantic schemas (URL patterns, connection distance pattern ^[FSO]$, etc.).

Configuration/Dependencies​

  • Required runtime dependencies:
    • requests, pydash, naas_abi_core (Integration base, cache, storage utils), naas_abi_marketplace (ABIModule)
  • Optional dependencies (only used by specific paths):
    • pandas (Excel export)
    • langchain_core, pydantic (for as_tools)
  • Authentication:
    • Must provide valid LinkedIn cookies: li_at and JSESSIONID.
  • Storage:
    • Uses StorageUtils bound to ABIModule.get_instance().engine.services.object_storage.
    • Writes under LinkedInIntegrationConfiguration.datastore_path.

Usage​

Minimal client usage​

from naas_abi_marketplace.applications.linkedin.integrations.LinkedInIntegration import (
    LinkedInIntegration,
    LinkedInIntegrationConfiguration,
)

cfg = LinkedInIntegrationConfiguration(
    li_at="YOUR_LI_AT",
    JSESSIONID='"YOUR_JSESSIONID"',  # quotes are stripped automatically
    linkedin_url="https://www.linkedin.com/in/someone/",
)

li = LinkedInIntegration(cfg)

org = li.get_organization_info("https://www.linkedin.com/company/naas-ai/", return_cleaned_json=True)
profile = li.get_profile_top_card("https://www.linkedin.com/in/someone/", return_cleaned_json=True)
print(org.keys(), profile.keys())

LangChain tools​

from naas_abi_marketplace.applications.linkedin.integrations.LinkedInIntegration import (
    LinkedInIntegrationConfiguration,
    as_tools,
)

cfg = LinkedInIntegrationConfiguration(
    li_at="YOUR_LI_AT",
    JSESSIONID="YOUR_JSESSIONID",
    linkedin_url="https://www.linkedin.com/in/someone/",
)

tools = as_tools(cfg)
# tools is a list[StructuredTool]

Caveats​

  • Uses LinkedIn private endpoints (voyager + hard-coded GraphQL queryIds); these may break if LinkedIn changes/rotates query IDs.
  • LinkedInIntegration.__init__ performs a live call to resolve profile_public_id and will raise if it fails.
  • _make_request is cached for 1 day keyed by method + cookies + endpoint (+ params); responses may be stale within TTL.
  • get_profile_posts_feed contains assert statements expecting specific response structure; it can raise AssertionError on unexpected API responses.
  • get_post_reactions, get_post_comments, and get_post_reposts call _make_request with unsupported keyword arguments (prefix, filename) in this file; as written, this will raise TypeError at runtime if those code paths are executed.
  • clean_json removes keys containing "urn" anywhere in the key name; this may drop fields you care about.
  • Location filtering in search_people only maps "France"; other locations are ignored (with a print).