India publishes more court data in public than most jurisdictions of comparable size, and a developer who sits down expecting to consume it as an API will spend a frustrating fortnight discovering that publication and programmatic access are entirely different things. Almost everything official is a human interface. Everything programmatic is either commercial or someone's scraper.
That is not a complaint about the judiciary. The eCourts programme, the National Judicial Data Grid and the individual court websites have made the Indian court record vastly more accessible than it was fifteen years ago, and they did it for the litigant and the advocate rather than for the integrator. The gap this piece describes is a design consequence, not a scandal.
What follows is a survey of the free routes: what the official public surfaces actually contain, what they will and will not let you do, where free commercial tiers fit, and the one distinction that determines whether your side project ships or quietly dies. A one time credit grant funds a demo. Only a recurring free allowance funds a project.
The Official Surfaces, and What Is Genuinely There
Start with what the state publishes, because it is the source everything else derives from and because a great deal of confused vendor marketing exists to obscure that fact.
The eCourts services portal and CNR lookup
The eCourts services portal is the closest thing India has to a unified case status system. It covers District Courts and, through connected services, a great deal of the High Court estate. Its most useful single feature for anyone thinking about identity is CNR lookup: a matter carries a unique case identifier, and given that identifier you can pull the case status, the party names, the listing history and the orders that have been uploaded against it.
This is genuinely record level access and it is free. It is also a web form. You get one matter at a time, through a page designed for a person who knows which matter they want. There is no documented public API contract behind it that you may rely on, no versioning, no stability guarantee, and access controls including CAPTCHAs that exist precisely because automated retrieval is not the intended use. Treat it as the authoritative place to confirm a matter, which it is, rather than as a data source you can build a service on top of, which it is not.
The National Judicial Data Grid
NJDG is the statistical layer, and it is frequently misunderstood by people planning products. It publishes pendency and disposal statistics: how many cases are pending in a given state, district or court, broken down by age band, by case type, by civil and criminal, by how long they have been waiting. It is excellent for what it is, and what it is is aggregate.
NJDG is not a route to record level data. You cannot ask it for the matters behind a number. If your project is a pendency dashboard, a judicial statistics visualisation, a research piece on disposal rates across High Courts, or a policy argument about how long commercial suits take under the Commercial Courts Act 2015, NJDG is the right and free source. If your project needs the cases themselves, it will not help you no matter how long you stare at it.
Supreme Court and High Court websites
The Supreme Court and each of the 25 High Courts run their own websites, with their own judgment search pages, their own cause list sections, their own case status forms and their own conventions. Many publish daily orders. Most publish cause lists the evening before. Judgment search is usually available by date range, by bench, by party name, sometimes by act and section.
These are the richest free surfaces for judgments specifically, and they are also the most fragmented. Twenty five High Courts means twenty five interfaces, twenty five URL schemes, twenty five sets of case type abbreviations, and twenty five independent decisions about when to redesign. A scraper written against a High Court site is a maintenance commitment with an indefinite term and no deprecation notice.
Tribunal portals
NCLT, NCLAT, ITAT, CESTAT, SAT, TDSAT and DRT each publish orders and cause lists on their own sites, at their own cadence, in their own formats. If your work touches insolvency under the IBC 2016, tax appeals, securities or telecom disputes, this is where that record lives, and it is largely outside the eCourts fabric. Free to read. Fragmented to consume.
The distinction that matters most
A public website is not a public API. Publication means a person may read the record. It does not imply a stable machine contract, a rate policy, an uptime commitment, or permission to retrieve in bulk. Every one of these surfaces can change its markup on a Tuesday without telling anyone, and none of them owes your project anything. Read each site's terms of use before you point a script at it, and treat the registry as the authority on anything you intend to act on.
Why Rolling Your Own Is a Project Rather Than a Weekend
The obvious plan, when you discover the data is public and the API is not, is to scrape it. Plenty of people have. It is worth being clear about what that commits you to, because the first weekend goes well and the next two years do not.
- Interface churn. Court sites are redesigned without notice or migration path. Every redesign silently breaks the parser, and silence is the failure mode: your pipeline keeps running and quietly returns nothing, or worse, returns wrong fields mapped to right names.
- Access controls. CAPTCHAs and similar measures exist on several surfaces. They are there deliberately. Defeating them is not a technical detail you can shrug at, and it changes the character of what you are doing from reading a public page to circumventing a control.
- Case type vocabulary. CRL.A., W.P.(C), CS(COMM), O.M.P., SLP and Arb.P. do not mean the same thing in the same way across every registry, and the abbreviations vary. Normalising them across courts is real, unglamorous work that never appears in the estimate.
- Uneven metadata. Fields present on a Supreme Court record are frequently absent on a District Court one, and recent records are richer than older ones. Any schema you design against one court will be wrong for another.
- Documents are the hard half. Getting a case status row is easy compared with getting the PDF, extracting usable text from a scanned order, and dealing with judgments recovered from print where character recognition is imperfect.
- Nobody owns it after you. A scraper is institutional debt. When the person who wrote it moves on, the team inherits a fragile dependency on twenty five external websites and no documentation of which quirks were deliberate.
The honest summary is that scraping court data is a legitimate approach for a research project with a bounded question and a fixed end date, and a poor foundation for anything with users who expect it to work next quarter.
The Two Shapes of Commercial Free Tier
This brings us to the commercial side, and to the distinction the entire piece turns on. Free tiers on data APIs come in two shapes that look identical on a pricing page and behave completely differently once you build against them.
The one time credit grant
You sign up and receive a fixed pot of credit. It is often generous, because generosity costs the vendor nothing once it is spent. You build, you test, you demo, and then one afternoon the pot is empty and every call returns an allowance error. The project is now a paid project or a dead project, and that decision arrives at whatever moment your usage happened to cross the line.
The recurring free allowance
You receive an amount of credit that refreshes on a cycle. It is usually smaller. It is also permanent, which changes everything, because a hobbyist with a modest monthly allowance can run a real thing indefinitely: a personal case tracker, a research tool for their own practice, a bot that watches a handful of matters. The ceiling constrains scale, not existence.
The second shape is the only one that supports a side project, and the reason is structural rather than a matter of size. A side project's defining property is that nobody is paid to maintain it. It survives because the marginal cost of it continuing to run is zero. A one time grant sets a date, unknown in advance, on which that marginal cost becomes non-zero and someone has to make a purchasing decision about a thing they built for fun. Most people, at that moment, do not make it.
A one time credit grant is a trial dressed as a free tier. It funds the demo and defunds the project, and it does so on a date nobody chose.
Be equally clear about the other side. Free tiers are marketing. They exist to get you far enough into an integration that switching costs are real. That is a perfectly reasonable bargain and you should take it with open eyes: read what happens when the allowance is exhausted, read whether unused allowance expires, read what the paid rate is, and read whether the free tier's terms permit the thing you are actually planning to do. If a free tier's terms say non-commercial, evaluation only, and your side project has three paying users, you have a problem that no amount of credit fixes.
The Free Routes, Compared Honestly
Here is the whole landscape in one place. Read the stability column carefully, because it is the one that decides whether a thing you build keeps working.
| Route | Coverage | Record level access | Stability | Terms | Cost |
|---|---|---|---|---|---|
| eCourts services portal and CNR lookup | District Courts broadly, plus connected High Court services | Yes, one matter at a time by identifier or by party and case number | A human interface with no published machine contract, subject to access controls | Read the portal terms; automated bulk retrieval is not the intended use | Free |
| NJDG dashboards | National, statewise, district and court level pendency and disposal | No, aggregate statistics only | Reasonably stable as a published dashboard, but it is a dashboard | Published for public consultation and policy use | Free |
| Supreme Court website | Supreme Court judgments, orders, cause lists, case status | Yes, through the site's own search and status forms | Its own interface, its own conventions, redesigned on its own schedule | Site terms of use apply | Free |
| High Court websites | Each of the 25 High Courts, independently | Yes, per court, with per court search behaviour | Twenty five separate surfaces changing independently, the least stable option to build on | Twenty five separate sets of terms | Free |
| Tribunal portals | NCLT, NCLAT, ITAT, CESTAT, SAT, TDSAT, DRT and others | Yes, per tribunal, formats vary widely | Fragmented, largely outside the eCourts fabric | Per portal terms | Free |
| Your own scraper over the above | Whatever you build and keep building | Yes, until the markup changes | You own every break, and breaks are silent | Bounded by each source's terms and by access controls you must not circumvent | Free in money, expensive in time |
| One time credit grant on a commercial API | Whatever the vendor covers | Yes, with a documented contract | Stable while credit lasts, then it stops | Vendor terms, often evaluation only | Free until exhausted |
| Recurring free tier on a commercial API | Whatever the vendor covers | Yes, with a documented contract | Stable and renewing, capped by the allowance | Vendor terms, check any non-commercial restriction | Free within the cap |
What You Can Genuinely Build for Free
Now the useful part. Constraints are not a reason to give up, they are a specification. A modest recurring allowance, combined with a rate ceiling of the kind commercial APIs apply, supports a real class of software. Here is what fits.
- A personal case tracker for a handful of matters. Ten to fifty matters, checked once or twice a day, is a few hundred calls a month. It is the single best fit for a free tier, and it is genuinely useful: your own matters, your own next dates, your own alerts, without opening seven portals.
- A research helper over a narrow slice. One court, one act, one bench, one year. Section 138 of the Negotiable Instruments Act 1881 in one High Court. Section 34 challenges under the Arbitration and Conciliation Act 1996 in one commercial division. Narrow enough that the query volume is bounded and the results are readable.
- A proof of concept for a pitch. Enough working software to show a partner, a client or an investor that the idea holds. Free tiers are excellent at this and it is the use they are designed for.
- A citation or identifier resolver. Paste a case number or a CNR, get the canonical record back. Low volume, high value, and it makes every other tool you own better.
- A judge name autocomplete. A substring search over a judge name list is cheap, deterministic and draws no AI credit. It is a small feature that removes a whole category of user error in any filter you build.
- A teaching tool or a student project. A clinic exercise, a dissertation dataset over a bounded slice, a classroom demonstration of how case types and identifiers work. Bounded scope, real data, no budget.
What You Cannot Build for Free, and Why That Is Not Meanness
Equally, be realistic about the ceiling. These are not arbitrary restrictions; each of them is a workload whose cost to serve is real and recurring.
The pattern across that list is that free tiers are sized for a person and priced for a person. The moment a workload scales with someone other than you, whether that is users, parties or documents, the economics change and honest pricing changes with them.
How to Evaluate a Free Tier Before You Build Against It
If you are going to spend a month of evenings on something, spend twenty minutes first on the terms. The questions below take one email to a vendor and save an unpleasant discovery later.
Recurring or one time
Ask directly whether the free allowance refreshes on a cycle or is granted once. If the answer is once, you are evaluating a trial, and you should plan the project on the assumption that it ends when the credit does.
What the allowance actually buys
Credits are not calls. On any API that runs models, a semantic search or an AI analysis consumes far more than a keyword search or a case fetch. Ask for the per-endpoint breakdown and compute your own monthly consumption against it rather than against the headline number.
The rate ceiling, separately
Throughput and spend are different constraints. A limit expressed as requests per minute per key shapes your architecture regardless of how much credit you hold, and a limit like ten per minute means you queue rather than burst.
What happens at exhaustion
Does the API return a clean, documented error you can handle and show a user, or does it degrade in some way you have to reverse engineer? A 403 on an exhausted AI allowance that you can catch and explain is a far better failure than a silent truncation.
The commercial restriction
Read whether the free tier permits commercial use. If your side project might one day have a paying user, know now whether that transition is a plan change or a terms violation.
Expiry and rollover
Unused allowance that lapses raises your effective cost the moment you start paying, and it shapes whether bursty usage is affordable. Ask before you build, not at renewal.
The exit
Can you export what you have stored, and are you permitted to keep it? A free tier you cannot leave is a switching cost the vendor has purchased with credit.
The failure that ends side projects
It is almost never a technical failure. It is a Tuesday when the grant runs out, the calls start returning an allowance error, and the person who built the thing for their own use decides it is not worth a purchase order. Check the shape of the free tier before you write the first line, because the shape decides whether the project has an end date you did not choose.
The Honest Boundary Between Free and Paid
Where does CourtMesh sit in this survey? Behind the API is a keyword corpus of roughly 310 million cases sourced directly from official government portals, the eCourts system, the NJDG and court registries, with no third party intermediary in between. Coverage spans the Supreme Court, all 25 High Courts, the district judiciary and tribunals including NCLT, NCLAT, ITAT, CESTAT, SAT, TDSAT and DRT. There are twelve endpoints, documented in the endpoint reference, and an MCP server for anyone wiring court data into an agent rather than into a web app.
The limits deserve the same plainness. The semantically embedded and AI analysed corpus is far smaller than the keyword corpus, in the low millions, so semantic search runs against an analysed slice rather than against everything indexed. Metadata completeness varies by court and by year: District Court records are thinner than Supreme Court records, and older records are thinner than recent ones. And the registry remains the authority. Anything aggregated, here or anywhere, is a view of what registries published, and it should be confirmed against the official record before you act on it.
That combination is the actual value of a paid API over free surfaces, and it is worth stating without dressing it up. You are not buying data that is secret. You are buying one contract instead of twenty five interfaces, normalised case types instead of registry specific abbreviations, documented error behaviour instead of a redesign you find out about from a user, and somebody else's responsibility for keeping it working. For a bounded research question, the free surfaces are entirely sufficient and you should use them. For anything that has to be working next quarter, you are choosing between paying a vendor and paying an engineer, and the engineer is not cheaper.
Start on the free routes, and know when you have outgrown them
The eCourts portal, NJDG and the court websites are real, free and authoritative, and for a bounded question they are the right answer. What they are not is a stable machine contract you can build a product on. The CourtMesh API gives you one documented interface over roughly 310 million cases from official government portals, across the Supreme Court, the 25 High Courts, District Courts and tribunals. Look at what the twelve endpoints actually return at the API documentation, check the tier structure and what happens when an allowance is exhausted at API pricing, and see the overall shape of the API at the API overview before you commit an evening to a scraper.
Explore CourtMesh
