How AI agents read new federal court opinions
An AI agent reads new federal opinions by calling a court wire at runtime. The wire carries CourtListener feeds for the Supreme Court and the Second, Ninth and Federal Circuits plus GovInfo, and returns each case name, the court that issued it, the publication date and a link to the opinion document.
/mcp. Nothing here needs an API key or a signup.
Court opinions are a hard retrieval problem for a language model. They publish in irregular bursts, they are long, and the case name alone tells you almost nothing about whether it matters to you. What an agent needs is not a legal database but a short answer to one question: has anything published since I last looked, and in which court.
The five feeds on this wire
| Court | Filter value | Publishes |
|---|---|---|
| Supreme Court | scotus | Opinions and orders during the Term, nothing during recess |
| Second Circuit | ca2 | Continuously, mostly summary orders |
| Ninth Circuit | ca9 | Continuously, the highest volume of the four |
| Federal Circuit | cafc | Patent, trade and veterans appeals |
| GovInfo | govinfo | US Courts opinions in irregular bursts |
That is four courts and one aggregator, not all 94 district courts and not all 13 circuits. If your matter sits in the Fifth Circuit or a district court in Texas, this wire will return nothing about it, and nothing is not the same as no ruling. The boundary is worth stating in the agent's own output rather than leaving the user to assume full coverage.
Recess is real, and an empty result is the correct answer
A Term of the Supreme Court begins on the first Monday in October and the Court typically recesses at the end of June. For roughly a quarter of the year, a query filtered to scotus returns zero opinions and that is accurate. An agent that treats zero as a failure will either retry pointlessly or, worse, fill the gap from training data and report a decision that did not happen this week.
The Justices do not stop working during recess. They evaluate roughly 130 petitions a week seeking review, which is why order lists resume before opinions do. So the honest shape of the answer changes by season: during the Term, "here is what published"; during recess, "nothing published in this window, and the Court is between sittings."
The same discipline applies to the circuits. A quiet Ninth Circuit day is common and does not mean the feed broke. Reporting an empty window as an empty window is the behaviour described in why agents hallucinate recent events.
The call an agent makes
# free preview, top 3 results, no payment and no wallet curl "https://thebotwire.com/court/preview?since=7d" # paid, $0.01: opinions from all five feeds in the last 7 days curl "https://thebotwire.com/court/opinions?since=7d&limit=20" # one court, one party or topic curl "https://thebotwire.com/court/opinions?type=cafc&q=patent&since=14d"
Parameters are type (the five values above), q (party name, topic or keyword), since (30m, 2h, 24h, 7d, 14d) and limit (1 to 50). This wire keeps a 14-day retention window and defaults to a 7-day lookback rather than 24 hours, because courts publish in bursts and a one-day window misses most of them. The route costs $0.01 per call in USDC on Base, with no API key and no signup.
An opinion is not a holding
The most expensive mistake a legal agent makes is not missing a case. It is summarising a newly published document as though it settled something.
- A summary order is not precedent. Much of what the Second Circuit publishes
is non-precedential by its own terms. The feed does not label this, so the agent must read the document before characterising it.
- A panel decision can be reheard. An opinion published today can be vacated
by an en banc grant later, and the wire has no way to know that yet.
- Case name matching is unreliable. Party names repeat across unrelated
matters. Match on the docket and the court, not on a string.
The safe pattern is narrow: report that a document published, name the court and the date, link the opinion, and stop. Let the human or a downstream retrieval step read the text. That keeps the agent inside what the wire actually knows.
For adjacent government signals, new rules arrive on the federal regulations wire and charges and settlements arrive on the DOJ enforcement wire. Which endpoint answers which question is listed in the routing table.
FAQ
Can an AI agent get Supreme Court opinions the day they publish?
Yes, during the Term. Opinions reach the CourtListener feed shortly after release and appear on the wire on the next ingest run. Between the end of June and the first Monday in October the Court is in recess, so the correct answer for a scotus query in August is that nothing published, not that the agent failed.
Does this cover district courts or state courts?
No. This wire carries the Supreme Court, the Second, Ninth and Federal Circuits, and the GovInfo US Courts feed. It does not carry district courts, other circuits, or any state court. Claiming otherwise would set an agent up to report silence as proof that no ruling exists.
Why not pull the CourtListener feeds directly?
You can, and for a single court that is a reasonable choice. The reason to use a wire is that a legal agent rarely stops at opinions: the same agent usually wants regulations, enforcement actions and filings too, and one call shape across 40 wires beats five parsers with five schemas and five failure modes.
How fresh is the data?
All five court feeds sit in the sparse ingest tier, polled at most every 15 minutes, because courts publish daily at most. The observed interval between ingest runs on this service is about 294 seconds and the last poll time is published live at /health.
Sources
Term dates and the petition volume on this page come from the Supreme Court of the United States, The Court and Its Procedures, which states that a Term begins on the first Monday in October, that the Court typically recesses at the end of June, and that the Justices evaluate approximately 130 petitions each week. Opinion data on this wire comes from CourtListener, maintained by the Free Law Project, and from GovInfo.
Related: How do AI agents track new US federal regulations? · How do AI agents monitor DOJ enforcement actions?