@memberjunction/actions-peopledatalabs

A table-agnostic People Data Labs client plus one thin Action. The client fetches and returns normalized person data; it never writes to the database, and it never returns emails, phone numbers or addresses. The Action exists for agents and workflows. Code that needs PDL data calls the client directly, never the Action.

What it covers

Credentials

Two paths, in order:

1. A CompanyID resolves that company's active People Data Labs row in MJ: Company Integrations, then its CredentialID in MJ: Credentials, whose Values hold { "apiKey": "…" }.
2. The PDL_API_KEY environment variable.

CompanyIntegration.APIKey is never read: it is not decrypt-on-read, so a synced value comes back as ciphertext and PDL answers 401. A credential that exists but is malformed throws rather than falling back to the environment.

The MJ: Integrations row named People Data Labs ships in metadata/integrations/.

Usage

import { PDLPersonClient } from '@memberjunction/actions-peopledatalabs';

const resolved = await PDLPersonClient.ForCompany(companyID, contextUser);
if (!resolved) throw new Error('no PDL key');
const result = await resolved.Client.BulkEnrich([
    { Key: personID, Email: 'someone@example.org' },
    { Key: otherID, FirstName: 'Ada', LastName: 'Lovelace', Company: 'analytical.example' },
], { MinLikelihood: 6 });
for (const match of result.Matches) {
    // match.Matched, match.PDLID, match.Likelihood (1..10), match.LinkedInURL,
    // match.Education[] { SchoolName, SchoolID, Degrees, Majors, StartDate, EndDate }
    // match.Experience[] { CompanyName, CompanyDomain, Title, StartDate, EndDate, IsCurrent }
}
Copy

Rules the client enforces:

• At most 100 inputs per call, and every input needs an identifying field: PDLID, Email, LinkedInURL, or first name plus last name plus company. Both checks run before any HTTP call.
• min_likelihood is sent inside each request's params. At the request level PDL silently ignores it and returns low-likelihood guesses.
• Results are correlated by position and checked against the echoed metadata.key. A mismatch throws rather than attaching one person's history to another.
• Per item: 200 is a match, 404 a miss (free, not an error), anything else sets ErrorMessage on that item only.
• Top level: 401 and 403 become PDLAuthError, 402 PDLCreditsExhaustedError, 429 PDLRateLimitError with RetryAfterSeconds from x-ratelimit-reset, anything else PDLRequestError. The client never sleeps.
• Dates arrive as YYYY, YYYY-MM or YYYY-MM-DD and are normalized to the first day of the year or month.
• Required and DataInclude are passed through when set, but PDL did not apply them on the plan this was built against. The field mapping here is the real minimization.

A note on schools: PDL may file a degree under the parent university. A Harvard Kennedy School master's came back as school.name = "harvard university" with major public administration. CleanSchool is not reliable alone (it mapped "John F. Kennedy School of Government" to john f. kennedy school); clean, then confirm with SuggestSchools, which returns ids and counts.

The Action

People Data Labs - Enrich People (DriverClass PDLEnrichPeopleAction), category Data Enrichment.

Result codes: SUCCESS, NO_MATCHES (a success), RATE_LIMITED, CREDITS_EXHAUSTED, CREDENTIALS_NOT_FOUND, CONFIGURATION_ERROR, VALIDATION_ERROR, ERROR.

Tests

pnpm test. Every test uses an injected fetch and the redacted fixtures under src/__tests__/fixtures/, captured from live responses on 2026-10-12. Nothing reaches PDL from a test.