The machine layer that makes INSC neutral citations the default over legacy SCC cites — for Claude, Cursor, and every drafting agent you run.
One open endpoint. Five courtroom-certain tools. Every ratio zero-hallucination, paragraph-anchored, and extractive — traceable to the official e-SCR record, never paraphrased.
Three copy-paste configurations. Each card's button copies the exact line to your clipboard — the next step is pasting it where it belongs.
Paste into .mcp.json under Claude Desktop's MCP server settings (Settings → Developer → Edit Config). Restart Claude Desktop — the five tools appear in the sidebar.
{"mcpServers": {
"insc-one": {
"url": "https://insc.one/mcp",
"transport": "streamable-http"
}
}}
Run it once in your terminal (user scope keeps it available in every project session). Your drafting agent picks up all five tools instantly.
claude mcp add --transport http insc-one https://insc.one/mcp
Fully offline or air-gapped drafting? The same five tools run locally as a stdio server from the open repo — no endpoint, no keys. Node 20+.
npx tsx mcp/index.ts
https://insc.one/mcp
https://insc.one/health
Five tools, one bench-aware citator behind each. Inputs are forgiving; outputs are courtroom-certain.
| Tool | What It Delivers | Sample I/O |
|---|---|---|
| resolve_citation | Any INSC, SCC, AIR, MANU, or party name → canonical INSC neutral citation, Universal Citation Rosetta parallels, official e-SCR link, and live treatment status. | in "(2014) 2 SCC 1" out "2013 INSC 768 · (2014) 2 SCC 1 · AIR 2014 SC 187 · e-SCR · good law" |
| get_ratio | Zero-hallucination extractive ratio decidendi, strictly paragraph-anchored — every holding maps to the verbatim judicial text it came from. | in "2013 INSC 768" out "Ratio @ paras 111–120 (verbatim, extractive)" |
| search_cases | Semantic and keyword search across 76 years of Supreme Court precedent — the bench-aware citator engine, not a keyword haystack. | in "arrest safeguards CCO" out "2014 INSC 492 (Arnesh Kumar) + 3 more" |
| check_treatment | Good law, overruled, or higher-bench signal — the Dawoodi Bohra bench-strength hierarchy, run automatically on every citation before your agent cites it. | in "1996 INSC 547" out "Superseded — see 5-Judge Constitution Bench, 2005" |
| format_citation | Paste-ready, INSC-primary citation strings with parallel reporters and official links — drop straight into pleadings, notes, and client briefs. | in "2013 INSC 768" out "Lalita Kumari v. Govt. of U.P., 2013 INSC 768 (= (2014) 2 SCC 1)" |
This is exactly what your drafting agent does once the connector is in.
The same machine layer, machine-readable, for every agent on the market — GPT, Gemini, and local models included.
The full machine-readable INSC corpus — one structured record per case: neutral citation, parallel reporters, treatment flags, paragraph anchors.
The complete citator context in a single pull — engine, tools, citation conventions — for agents that work from document context.
The lightweight index: what insc.one is, what it does, and where to go for depth. The standard first fetch.
Written for lawyers who have never heard of MCP. Two minutes to understand what this is and what it does for you.
MCP is the open standard that lets AI apps — Claude, Cursor, Claude Code — ask external sources for verified answers instead of guessing. Think of it as a common port: the AI app plugs into your source, the source hands over the answer, both sides use the same connector. For you it means one practical thing: when an AI is drafting with your help, it can look a case up on insc.one and cite it correctly — instead of inventing a citation from memory.
No account. No API key. Nothing to install on your computer. The connector is open access at
https://insc.one/mcp — you only do anything on the
AI app's side (Claude Desktop → Connectors, or the one-line command on the "Install" cards above).
Two checks. First, open https://insc.one/health in any
browser — you should see {"ok":true,"cases":39,...}.
Second, in a connected AI app, ask: "Use the insc.one tools to give me the neutral citation and holding for Arnesh Kumar."
It should call the tools and answer with 2014 INSC 492 plus the paragraph-anchored ratio.
It says so. Every tool returns "not in the insc.one index" with a pointer to the official Supreme Court records — it will never fill the gap with a made-up citation. That honesty rule is the whole point: a tool that invents case names is worse than no tool.
The connector reads the same 39-case landmark corpus as the Citator Workbench — neutral citations, parallel reporters,
paragraph-anchored ratios, treatment status. It updates every time the citator data is rebuilt and deployed;
/health reports the case count so you can see it at a glance.
Read-only and stateless: the connector answers questions, stores nothing, and runs no sessions. No personal data, no drafting content, no credentials involved on either side.