Explore the labor market with Core LMI¶

Core LMI (also called Agnitio) adds employment, earnings, industry, education, and regional measures to the server. It uses your existing Lightcast credentials with the agnitio scope. Start with dataset discovery: the catalog reflects your contract.

During the November 12, 2026 review, the account exposed 52 datasets. Data queries were verified for occupation, industry, and education completions, including the examples below. The available releases included 2026.4. Those are observations from this review, not defaults or a promise that another account has the same access.

Questions you can ask¶

Core LMI measures and projections differ from online job advertisements. Use Job Postings tools for advertised demand, and Core LMI for the measures described by the selected dataset.

Discover → inspect → query¶

1. Call list_core_lmi_datasets with {"query":"occupation"}. Set include_definitions: true to also search descriptions. Follow pagination.next_offset if more entries remain.
2. Call get_core_lmi_versions with the returned dataset name. latest_numeric selects the newest numeric release regardless of the API’s array order. Special releases such as base and example require explicit selection.
3. Call get_core_lmi_metadata with that dataset and release. Inspect attributes, dimensions, and metrics. Filter metrics using metric_query, and page through metrics_pagination.
4. Call get_core_lmi_dimension to find exact codes. Choose search or hierarchy filters (level, aggregation_path) in a call. Preserve the returned child code exactly; different taxonomies use different code formats.
5. Call query_core_lmi, pinning the release you inspected.

For example, inspect available earnings metrics:

{
  "dataset": "emsi.us.occupation",
  "version": "2026.4",
  "metric_query": "Earnings.",
  "limit": 20
}

Find the occupation code with get_core_lmi_dimension:

{
  "dataset": "emsi.us.occupation",
  "version": "2026.4",
  "dimension": "Occupation",
  "search": "Software Developers"
}

The live response identified 15-1252. Area discovery identified California as 6 and Washington as 53. These are Core LMI codes, so use the discovered values instead of borrowing IDs from Titles or LOT.

Compare employment and annual earnings¶

Tool: query_core_lmi. This request returns one row per state. The worker group includes QCEW and non-QCEW employees; self-employed workers and extended proprietors are excluded by this explicit selection.

{
  "dataset": "emsi.us.occupation",
  "version": "2026.4",
  "query": {
    "metrics": [
      {"name": "Jobs.2025"},
      {"name": "Jobs.2030"},
      {"name": "Earnings.Median.Annual"}
    ],
    "constraints": [
      {"dimensionName": "Area", "map": {"California": ["6"], "Washington": ["53"]}},
      {"dimensionName": "Occupation", "map": {"Software Developers": ["15-1252"]}},
      {"dimensionName": "ClassOfWorker", "map": {"Employees": ["1", "2"]}}
    ],
    "sortBy": [{"name": "Area", "direction": "ascending"}],
    "limit": 10
  }
}

Interpret the units from metadata. Annual earnings, hourly earnings, total industry earnings, job counts, and advertised salaries answer different questions. Keep missing or suppressed values as missing; do not convert them to zero.

Build a custom regional industry view¶

This verified query combines two counties into one region. Each key under map becomes an output label, while its array defines the codes to aggregate.

{
  "dataset": "emsi.us.industry",
  "version": "2026.4",
  "query": {
    "metrics": [{"name": "Jobs.2025"}, {"name": "Establishments.2025"}],
    "constraints": [
      {"dimensionName": "Area", "map": {"King + Kitsap counties": ["53033", "53035"]}},
      {"dimensionName": "Industry", "map": {"Full-Service Restaurants": ["722511"]}},
      {"dimensionName": "ClassOfWorker", "map": {"Employees": ["1", "2"]}}
    ],
    "limit": 10
  }
}

For separate counties, use separate map labels. For a hierarchy expansion, the verified industry query also accepts {"dimensionName":"Area","mapLevel":{"level":3,"predicate":[53]}} to request Washington counties. Select small pages when expanding a hierarchy.

Investigate concentration and change¶

Keep the industry and worker constraints above, and use one area: {"King County, WA":["53033"]}. Replace metrics with:

[
  {
    "name": "Jobs.2025",
    "as": "Concentration",
    "operation": {"name": "LocationQuotient", "geoparent": "0", "along": "Industry"}
  },
  {
    "name": "Jobs.2025",
    "as": "Change factors",
    "operation": {"name": "ShiftShare", "geoparent": "0", "along": "Industry", "base": "Jobs.2020"}
  }
]

Both operations passed live testing. Shift share returns multiple component columns; preserve their returned labels rather than treating the operation as a single scalar. The national comparison is selected with geoparent: "0".

Examine education completions¶

This verified example requests computer-science bachelor’s completions at UW Seattle. Institution, program, and award codes were checked through dimension discovery.

{
  "dataset": "emsi.us.completers",
  "version": "2026.4",
  "query": {
    "metrics": [{"name": "Completions.2025"}],
    "constraints": [
      {"dimensionName": "Unit", "map": {"UW Seattle": ["236948"]}},
      {"dimensionName": "Program", "map": {"Computer Science": ["11.0701"]}},
      {"dimensionName": "AwardLevel", "map": {"Bachelors": ["5"]}}
    ],
    "limit": 10
  }
}

Use the dataset’s definitions when choosing between Completions and Completers. Award counts and counts of people are different measures.

Read and page results correctly¶

data is column-oriented. Each column has name, type, and rows; values at the same index form one record. The wrapper preserves upstream columns, extra response fields, errors, timings, and release provenance.

Queries return at most 500 rows per page (100 by default), with at most 20 metrics. This server exposes a bounded window of the first 5,000 rows. Narrow the dimensions for larger analyses.

Live testing found that the upstream service ignored offset and reported totalRows for the limited response. To prevent duplicate pages, the wrapper requests a prefix of offset + limit + 1 rows, sets the upstream offset to zero, and slices the requested page locally. The extra row detects whether more results exist. Later pages re-fetch the prefix and therefore cost more work than the first page.

• Keep the dataset, release, filters, and sortBy identical between pages. Use enough sorting dimensions to disambiguate rows.
• Set query.offset to pagination.next_offset; nonzero offsets require sortBy.
• pagination.returned is the delivered row count; fetched includes the prefix and lookahead.
• totalRows is preserved as upstream metadata. pagination.total is null because it is not an authoritative count of all matching rows.
• If has_more is true but next_offset is null, the 5,000-row window is exhausted. Narrow the query.
• Metadata pagination happens locally too; hierarchy filters reduce the upstream metadata response.

Known service limitation¶

The account can discover emsi.us.costoflivingindex, its COLI metric, and area codes. Data requests using the documented map, mapLevel, and asIdentity forms returned HTTP 400 during this review. One response requested mapChildren, which is absent from the current published query schema. This integration does not invent that undocumented contract or claim cost-of-living data queries passed. Resolve this discrepancy with Lightcast before relying on that dataset’s query results.

Re-run the live checks¶

uv run python -m mcp_lightcast.diagnostics --core-lmi --output test-results/core-lmi-live-results.json

This explicitly uses your credentials and consumes quota. The report contains outcomes, release metadata, and the accessible dataset count, without credentials or market values. Core LMI requests are spaced at one per second by default, consistent with the documented 300 requests per five minutes.

Official references¶

Service overview and quota, dataset catalog, version metadata, hierarchies, dimension search, and query schema.