Midpage

What it is

Midpage is a US legal research platform built on one commitment: every answer stays clickable back to the primary source it came from. Litigators use it to research case law, check citations before a brief goes out, pull dockets from PACER, and draft against opinions the system actually retrieved rather than recalled. Founded in 2022 and run from New York, it raised a $4M seed in June 2025 and reports 300+ law firms and 10,000+ litigators on the platform.

The distinguishing choice is distribution, not the interface. Midpage sells the same corpus three ways: its own web app, connectors that put the data inside Claude, ChatGPT and Perplexity, and a raw data licence — API, MCP server, or a dedicated PostgreSQL replica — for other legal-tech vendors to build on. CEO Otto von Zastrow’s stated reason is that lawyers want fewer surfaces, not another one. The evidence that the bet landed: by July 2026, usage through the Claude connector was running at roughly twice the volume of Midpage’s own web interface.

July 2026 is when it stopped being a case-law tool. On 23 July 2026 Midpage added statutes, regulations, constitutions, executive orders and agency guidance for all 50 states, DC and the federal government — 4M+ documents in a single release. The site now claims 16M+ opinions and 6M+ statutory and regulatory documents, with agency guidance from the IRS, SEC, USPTO, DOL, CFPB, EEOC, FDA, HUD, OFAC and USCIS. Every statute and regulation is versioned, so you can pull the text as it read on the date your facts arose rather than the text as it reads today.

Why it shows up in Legal Ops stacks

• The citator runs on a graph, not an editorial team. Midpage tracks 111M+ links between opinions and derives treatment signals — negative, caution, neutral — from that graph. Cases are ingested multiple times a day; statutes and regulations refresh weekly. Westlaw and Lexis get the same result by paying people to read every new decision, which is why their prices look the way they do.
• It is the cheapest published price in AI legal research, by a factor. $100 per user per month against roughly $428 for Westlaw Precision with CoCounsel and $499 list for Paxton. Alexi, CoCounsel and vLex Vincent AI publish no price at all.
• The MCP connector is the actual product for most buyers. Claude does the reasoning and the drafting; Midpage supplies retrieval, the citator signal and the hyperlink back to the opinion. Nobody learns a second research UI, and the citation check happens before the text reaches the brief. See best MCP tools for legal ops.
• Other vendors’ AI already runs on it. Litera, Noxtua, Perplexity, BriefCatch, GC AI and the US Courts license the underlying data, and Midpage reports 200K+ weekly API and MCP calls from partners. A research tool that other research tools buy from is a different risk profile than a thin wrapper.
• US law on European infrastructure exists now. The Noxtua partnership announced 29 April 2026 puts Midpage’s US corpus behind Noxtua’s European hosting, so a DACH firm can research US authority without the client data leaving EU servers.

Pricing reality

Professional is $100 per user per month, published, monthly or annual. It includes the full corpus, the AI citator, agentic PACER access, team billing and SSO, and “unlimited” AI research usage with an explicit suspension right for abnormal volume. The free trial takes no card. Law students and faculty pay $25 per user per month with proof of enrollment.

Two costs sit outside the seat. PACER retrieval is pass-through — Professional includes $25 a month in credits, and past that you pay the court’s own per-page fees. Historical point-in-time versions of statutes and regulations sit behind a separate entitlement, so confirm it is in your quote if litigating conduct from prior years is the reason you are buying. API and bulk data access are quote-only and scoped on volume.

The number that decides most evaluations: five litigators on Midpage costs about $6,000 a year, roughly what one Westlaw Precision with CoCounsel seat costs. Security is SOC 2 Type II and HIPAA audited, with zero-day retention required from its model providers.

Best for

Litigation boutiques, plaintiff-side firms and solo litigators who need federal and state case law with defensible citations, already work inside Claude or ChatGPT, and cannot justify a Westlaw or Lexis contract. It is also the cheapest way for an in-house team to give an existing Claude deployment real US primary law instead of model recall.

Alternatives and when to pick them

• Thomson Reuters CoCounsel with Westlaw — the market-share incumbent. Pick it when you need secondary sources (practice notes, treatises, forms) or when KeyCite is the standard your opposing counsel, your court and your malpractice carrier recognize by name.
• LexisNexis Protégé — the other incumbent. Pick it when Shepard’s, Lexis secondary content and a firm-wide enterprise agreement already exist and the marginal cost of the AI layer is small.
• vLex Vincent AI — pick it when the matter crosses borders. vLex covers dozens of jurisdictions outside the US; Midpage covers none.
• Harvey — the fastest-growing entrant in legal AI, at $300M raised in its most recent round. Pick it when the job is firm-wide transactional and advisory work, not litigation research.
• Alexi — pick it when you want finished research memos produced for you rather than a corpus your own agent queries.

If your practice is non-US, or a standing order or carrier requirement names KeyCite or Shepard’s specifically, Midpage does not replace the incumbent. Run it alongside for a year and let the seat count decide the renewal.

Watch-outs

• US-only, and four state codes were still incomplete at launch. Georgia, Tennessee, Mississippi and Arkansas codes were being reconstructed from session laws as of July 2026, because those states publish through Lexis. Guard: take the primary authorities from your last ten research memos and confirm each one resolves in Midpage before you cancel anything.
• The citator is AI-derived and has no published independent audit. Vendor accuracy claims for treatment signals are self-reported. This is not a Midpage-specific failing — a 2018 Law Library Journal study by Paul Hellyer found Shepard’s, KeyCite and BCite agreed on negative treatment in only 53 of 357 sampled relationships. Guard: spot-check negative treatment against your incumbent citator on every authority you actually cite, and keep that step in the workflow for the first two matters.
• “Unlimited” carries a suspension clause. Midpage reserves the right to suspend abnormally high usage, and an agentic workflow inside Claude generates far more queries than a person typing searches. Guard: run a two-week trial at your real agent volume and get the threshold named in the order form before annual commit.
• Your research surface is partly Anthropic’s release schedule. Connector behaviour changes when the host platform changes, and the vendor itself has acknowledged early bugs in the Claude Desktop integration. Guard: keep the web app in the workflow as the fallback path, and confirm the same query returns the same authorities in both surfaces during the trial.
• This is a small vendor competing with Thomson Reuters and RELX. $4M seed money against two incumbents with editorial operations built over a century. Guard: negotiate a data-export and API-continuity clause at signature, and price the switch back before you let the Westlaw contract lapse.

For the surrounding category, see best legal AI research tools and grounding vs hallucination in legal AI.