CourtListener Case Law Scraper — US Opinions, Dockets & Oral Arguments

Search CourtListener's corpus of US case law: case name, court, docket number, judges, citations and cite counts, filing dates, and the opening text of each opinion. Also covers RECAP dockets, oral arguments and judges.

Please use a free API token

CourtListener is run by the Free Law Project, a non-profit, and its API is rate-limited on purpose. Measured: 5 requests a minute anonymously, then a 429 with an honest Retry-After. That is about 100 results a minute at best.

A token is free from courtlistener.com and lifts the limit substantially. Without one this actor deliberately paces itself to one request every 13 seconds and says so in the log. It is not trying to be slow; it is trying not to hammer a charity's servers, and no proxy setting changes that — the limit is a published rate limit, not an anti-bot measure.

Input

Field	Type	Default	Description
searchQueries	array	required	Same syntax as CourtListener's own search box
searchType	string	opinions	opinions, recap, oral-arguments, people
courts	array	—	Court IDs, e.g. scotus, ca9, dcd
filedAfter / filedBefore	string	—	YYYY-MM-DD
maxItemsPerQuery	integer	100	20 results per request
apiToken	string (secret)	—	Free from courtlistener.com — see above

Output

{
  "_input": "copyright fair use",
  "_source": "S1-courtlistener-api",
  "_scrapedAt": "2026-09-09T13:22:07Z",

  "clusterId": 9231237, "docketId": 65678901,
  "url": "https://www.courtlistener.com/opinion/9231237/fort-bend-cnty-v-davis/",

  "caseName": "Fort Bend Cnty. v. Davis",
  "caseNameFull": "Fort Bend County, Texas v. Lois M. Davis",
  "docketNumber": "No. 18-525",
  "status": "Published",

  "court": "Supreme Court of the United States",
  "courtId": "scotus",
  "courtCitationString": "SCOTUS",

  "dateFiled": "2019-06-03", "dateArgued": "2019-04-22",
  "judge": "Ginsburg",
  "attorney": "Benjamin J. Horwich, San Francisco, CA, argued the cause for…",

  "citations": ["587 U.S. 541", "139 S. Ct. 1843"],
  "citeCount": 398,
  "lexisCite": "…", "scdbId": "…",

  "opinionCount": 1,
  "opinions": [{
    "opinionId": 9231238, "type": "combined-opinion",
    "authorId": 8589, "perCuriam": false,
    "downloadUrl": "…", "localPath": "pdf/…", "citesCount": 41,
    "snippet": "Justice GINSBURG delivered the opinion of the Court…"
  }],
  "snippet": "Justice GINSBURG delivered the opinion of the Court…",

  "searchType": "opinions"
}

Two things worth knowing

Cursor pages overlap slightly. Paging by the next cursor, page 2 repeated one of page 1's twenty results in testing. The actor tracks cluster IDs and emits each case once, so a run of N requests can return slightly fewer than 20 × N rows — that is correct, not a shortfall.

Several fields are real but sparse. posture, proceduralHistory, syllabus, panelNames, panelIds, neutralCite, dateReargued and dateReargumentDenied are genuine CourtListener fields that are empty for most cases; suitNature and courtJurisdiction populate for a minority (4 and 1 of 20 in a mixed-court sample, and none at all in a Supreme Court sample). They are shipped rather than hidden, because when they are present they matter — but an empty value here means the corpus has none, not that the scrape failed.

Full opinion text is not fetched. Each opinion carries a snippet — the opening of the text — plus downloadUrl and localPath pointing at the source document. Retrieving complete opinions would mean one request per case against the same rate limit, so it is deliberately out of scope.

Errors

_error	Meaning
invalid_input	Empty search query
no_results	The search ran and matched nothing
unexpected_shape	A 200 without a results list
blocked	Every TLS profile was refused
network_error	The ladder never reached the server

A rejected API token fails the whole run immediately rather than repeating the mistake for every remaining query — nothing downstream can succeed with bad credentials.