software.nasa.gov

lookup-software-contract

Installation

Adds this website's skill for your agents

 

Summary

Look up software in NASA's public Software Catalog by free-text keyword or by canonical case number, returning structured records with title, description, NASA field center, category, release type, version, dates, and download URL.

SKILL.md
237 lines

NASA Software Catalog Lookup

Purpose

Look up software in NASA's public Software Catalog (software.nasa.gov) — either by free-text keyword (returning a ranked list of matching titles) or by canonical NASA case number (returning the full record for a single piece of software). For each hit, returns: case number, title, marketing/technical descriptions, NASA field center, category, release type, version, dates, and any external download URL. Read-only; never submits the "Request Software" form.

When to Use

  • "Find NASA software for X" — keyword search across the full agency catalog.
  • "Get details on NASA software LAR-18744-1" — direct lookup by case number.
  • Programmatic monitoring for newly released or revised NASA software.
  • Cataloging / filtering by NASA field center (Ames, JPL, Goddard, Langley, etc.) or by software category (aeronautics, propulsion, data and image processing, etc.).
  • Anywhere you would otherwise scrape the rendered HTML — the JSON API is faster, structurally richer (24 fields per record vs. 4 in the rendered card), and has no auth, cookies, or anti-bot.

Workflow

The software.nasa.gov front-end is a thin client over a public Elasticsearch-backed JSON endpoint at /searchapi/multi/{center}/software/{offset}/9/{query}. No auth, no cookies, no rate-limit headers, no anti-bot Verified — plain HTTPS GET works from any IP. Page size is hard-coded to 9. Lead with the API path; the browser path is a fallback only.

A. Search by keyword

  1. Get the total result count (one HTML round-trip — the API itself does not return a count).

    GET https://software.nasa.gov/search/multi/{center}/software/9/{query}
    • {center} — one of: aw (agency-wide / all), arc (Ames), dfrc (Armstrong), grc (Glenn), gsfc (Goddard), jpl (JPL), jsc (Johnson), ksc (Kennedy), larc (Langley), msfc (Marshall), ssc (Stennis).
    • {query} — URL-encoded keyword(s). Spaces → %20. Hyphens allowed.

    Parse total from one of (both present in the HTML):

    • <div class="searchCount">(\d+) Search Results for "..."</div>
    • Inline JS literal let total=<N>;

  2. Fetch matching records via the JSON API, paginating by offset = 0, 9, 18, … until the API returns [] or offset ≥ total.

    GET https://software.nasa.gov/searchapi/multi/{center}/software/{offset}/9/{query}

    Returns a JSON array of Elasticsearch hits:

    [
  {
    "_index": "t2pd.software",
    "_id": "544057a42841f54dacba7027",
    "_score": 1,
    "_source": { /* full software record — see "Expected Output" */ }
  },
  ...
]

    The 24 fields in _source (e.g. case_number, marketing_title, marketing_desc, tech_title, sw_desc, category, center, release_type, sw_version, sw_url, release_aisc_date, revision_date) are the canonical record and are richer than the rendered card or the public detail page — you usually do not need step 3.

  3. (Optional) Fetch the public detail page for a single hit if you need the rendered "Contact Us" email, the FAQ-formatted release-type explainer, or a visual screenshot:

    GET https://software.nasa.gov/software/{case_number}

    Returns server-rendered HTML; the <h1> is {marketing_title}({case_number}) and labeled rows under "Software Details" carry Category, Reference Number, Release Type, and Operating System. The detail page is strictly a subset of the JSON record — there is no information here the API does not already have.

B. Direct lookup by case number

If the user already knows the canonical case number (e.g. LAR-18744-1, NPO-50498-1, GSC-15016-1), do not use the search API — it does a fuzzy multi-field match on the case-number string and returns center-prefix-mates ranked above the exact match (e.g. searching LAR-18744-1 returns LAR-19278-1 as the top hit, not the exact match). Instead:

  1. Hit the detail URL directly:

    GET https://software.nasa.gov/software/{case_number}
    • HTTP 200 with <h1>...({case_number})</h1> → exists. Parse <h1> and the "Software Details" rows for category, release type, etc.
    • HTTP 404 → unknown case number.

  2. (Optional) Backfill the rich JSON record by extracting the marketing title from the <h1> and querying the search API with that title — the exact-title query lands the canonical record as hit #1 with all 24 _source fields.

C. Autocomplete (optional)

For "did you mean" suggestions on a partial query (≥ 4 chars), call:

GET https://software.nasa.gov/searchsuggestions/{partial}

Returns {"success": true, "data": ["trajectory", "trajectory optimization", ...]} (up to 10 strings) or [] if no matches. Useful for query expansion before the main search call. Returns [] for queries shorter than 3-4 characters.

Browser fallback

Use only when the JSON endpoint is unreachable (sanctions, network policy, or — extremely rare — the Drupal cache is misbehaving). The public search page https://software.nasa.gov/search/multi/{center}/software/9/{query} is fully server-side rendered for the first 9 hits (Drupal Views) — no JS execution required to read them. Extract:

  • Per-result <div class="result"> block, each containing:
    • Case number: <a href="https://software.nasa.gov/software/([A-Z0-9\-]+)"> (also the anchor text)
    • Title: <div class="title">([^<]+)</div> (the marketing_title)
    • Description: <div class="description">([^<]+)</div> (the marketing_desc, truncated)
    • Release type: <div class="category">([^<]+)</div> (re-uses the category class but carries the release_type text — confusingly named)
  • Total count: <div class="searchCount">(\d+) Search Results (also let total=<N>; in the in-page JS).
  • Pagination via browser requires running the viewMore JS (document.querySelector(".viewMore").click() per +9 batch) — much more expensive than just hitting the JSON endpoint at the next offset. Skip the browser entirely once you have the count.

Site-Specific Gotchas

  • The "obvious" search-API URL is a trap. The view-more JS embedded in the search page advertises /searchapi/software/{query}/software/{page}/9/ — that endpoint always returns [] for every query, every offset, every center (confirmed across multiple iterations 2026-05-18). The actually-working endpoint is /searchapi/multi/{center}/software/{offset}/9/{query} — note the segment order: {offset}/9/{query} is correct, not {offset}/{query} (the trailing /9/ is the page size and must be present — without it the query parameter is silently ignored and you get the unfiltered agency-wide pager). This was the highest-cost discovery during scaffolding; the URL inside the page's JS is stale.
  • Query parameter is silently dropped if path shape is wrong. Any malformed variant of the path (missing /9/, wrong segment order, using aw in the query slot, etc.) returns the unfiltered top-9 of the catalog with HTTP 200 — there is no error signal. Always verify by submitting a query you know is unique (e.g. TPSSizer) and confirming the response shape matches expectations (hit count should be small for unique terms).
  • No total-count in the JSON. The API only returns the slice of 9; the total is published in the rendered HTML page (class="searchCount" and inline let total=<N>;). Either fetch the HTML once at offset=0 to grab the count, or paginate blindly until you get [].
  • Page size is hard-coded to 9. The /9/ in the path is the size and is not overridable — /10/, /100/, etc. all return {} (empty object, not an array) or []. Plan pagination in increments of 9.
  • Empty / very-short queries silently return browse-all-9. q="", q=" ", q="x" all return 9 results sorted by the catalog's default order (ARC entries first), not an error. Validate the query is ≥ 2 characters before issuing the request and treat short-query responses as "browse" rather than "search-result".
  • Direct case-number lookup via the search API is unreliable. Querying case_number strings does a fuzzy match on the prefix and ranks center-prefix-mates above the exact case (e.g. LAR-18744-1 returns LAR-19278-1 as hit #1). For exact-id lookup, fetch https://software.nasa.gov/software/{case_number} directly (HTTP 200 = exists, 404 = unknown).
  • The category field on result cards is actually release_type. In the rendered HTML (<div class="category">) the value is the release-type enum (Open Source, General Public Release, U.S. Release Only, U.S. and Foreign Release, U.S. Government Purpose Release), not the technical category. The actual technical category lives in _source.category (lowercase) / _source.sw_category (title-case display). Don't conflate them.
  • Field-center enum is closed. Valid {center} values: aw, arc, dfrc, grc, gsfc, jpl, jsc, ksc, larc, msfc, ssc. Unknown values return []. JPL's _source.center is JPL even though its case-numbers are NPO-* (legacy Jet Propulsion lab "Naval Propellant Office" prefix), not JPL-*. Don't try to infer center from case-number prefix.
  • Category-browse pages are not server-rendered. https://software.nasa.gov/software/category/{slug}/{center}/{page}/ returns an empty results div — these pages depend on JS to fetch and render their hits. There is no public JSON endpoint for category-browse (only the keyword /searchapi/multi/.../{query} endpoint, which does not accept a category filter). If you need category-scoped results, either: (a) call the keyword API with the category slug as the query (e.g. q=propulsion) — produces approximate results, or (b) call the keyword API for every term you care about and filter client-side on _source.category.
  • Latency / rate-limiting. Median response ~1.6-1.8s per API call from a US-West vantage point. No rate-limit headers and no formal block observed at 3 req/s; keep ≤ 2 req/s sustained to be polite.
  • CDN cache is aggressive. X-Drupal-Cache: HIT on most responses; Cache-Control: must-revalidate, no-cache, private is ignored by the CloudFront layer in practice. Newly-added software entries may take up to a few minutes to appear. The revision_date field in _source is the canonical "last modified" timestamp.
  • sw_url is sparsely populated. Open-source codes typically have a non-empty sw_url pointing to a GitHub/SourceForge/lab page; export-controlled codes (U.S. Release Only, U.S. Government Purpose Release) leave it blank and require the "Request Software" form (which this skill does not exercise).
  • sw_version_date: "1970-01-01T08:00:00.000Z" is a sentinel for "no version date". Treat as null. (The 8:00 offset is PST epoch-zero — a Drupal default.)
  • Browser session not always reachable. Driving software.nasa.gov through a Browserbase remote session requires resolving connect.{region}.browserbase.com (WSS/CDP endpoint), which is firewalled in some agent runtime environments — only api.browserbase.com (REST gateway, used by browse cloud fetch/browse cloud search) is reachable. This is fine for this skill since the JSON API path needs only HTTPS GETs — but it means automated screenshots may not be available depending on your sandbox.

Expected Output

A. Keyword-search response shape

One object per matching software. Pull the entire _source block — there are no positional arrays or decode tables to worry about; every field is named.

{
  "query": "trajectory",
  "center": "aw",
  "total": 53,
  "results": [
    {
      "case_number": "LAR-18744-1",
      "marketing_title": "Low Fidelity Space Systems Analysis Tools-Heliocentric Trajectory Tool",
      "marketing_desc": "This tool is produced is take advantage of pre-existing resources of known lambert trajectory solutions to various bodies, NEA's and more. ...",
      "tech_title": "Low Fidelity Space Systems Analysis Tools-Heliocentric Trajectory Tool",
      "sw_desc": "Full technical description (often longer than marketing_desc)...",
      "category": "design and integration tools",
      "sw_category": "Design and Integration Tools",
      "center": "LARC",
      "release_type": "U.S. Release Only",
      "sw_version": "1.0",
      "sw_version_date": "1970-01-01T08:00:00.000Z",
      "release_aisc_date": "2016-07-14T07:00:00.000Z",
      "revision_date": "2016-07-14T07:00:00.000Z",
      "sw_url": "",
      "sw_operating_system": "",
      "reference_number": "LAR-18744-1",
      "client_record_id": "LAR-18744-1",
      "id": "software_release_LAR-18744-1",
      "aisc_keyword": "",
      "catalog_note": "",
      "more_sw_info": "",
      "search": "(deduplicated full-text blob)",
      "title": "Low Fidelity Space Systems Analysis Tools-Heliocentric Trajectory Tool",
      "detail_url": "https://software.nasa.gov/software/LAR-18744-1"
    }
  ],
  "next_offset": 9
}

B. Direct case-number lookup response shape

{
  "case_number": "LAR-18744-1",
  "found": true,
  "title": "Low Fidelity Space Systems Analysis Tools-Heliocentric Trajectory Tool",
  "category": "Design and Integration Tools",
  "release_type": "U.S. Release Only",
  "operating_system": null,
  "detail_url": "https://software.nasa.gov/software/LAR-18744-1",
  "description": "Overview text from the rendered detail page...",
  "contact_email": "hq-dl-t2-ops-center@mail.nasa.gov"
}

C. Not-found shape (case-number direct lookup)

{
  "case_number": "ZZZ-99999-1",
  "found": false,
  "detail_url": "https://software.nasa.gov/software/ZZZ-99999-1",
  "http_status": 404
}

D. Empty-search shape (keyword)

{
  "query": "nonexistentxyz",
  "center": "aw",
  "total": 0,
  "results": [],
  "next_offset": null
}

E. Autocomplete-only shape (when partial query is too short to search)

{
  "partial": "traj",
  "suggestions": [
    "trajectory",
    "trajectory optimization",
    "trajectories",
    "trajectory model",
    "trajectory estimation"
  ]
}