Connect Cursor or Claude to Indian case law

What it is

The Model Context Protocol (MCP) is how AI clients such as Cursor and Claude call external tools. SooperKanoon runs an MCP server at https://sooperkanoon.com/mcp, so your assistant can look up judgments, bare act sections, and published advocate profiles directly instead of guessing from memory.

It reads the same index that powers sooperkanoon.com and uses the same API key and the same prepaid balance as our JSON REST API. There is no separate MCP signup, no second key, and no different price.

---

Quick start

---

Client configuration

The endpoint is a Streamable HTTP MCP server. Configure it as a remote (URL-based) server — there is nothing to install locally.

{
  "mcpServers": {
    "sooperkanoon": {
      "url": "https://sooperkanoon.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Replace YOUR_API_KEY with the GUID from your signup email. If your client cannot set an Authorization header, the legacy HTTP-GUID header works too:

HTTP-GUID: YOUR_API_KEY

A quick check from the command line — this lists the tools your key can call:

curl -X POST https://sooperkanoon.com/mcp \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

Authentication is required on every request, including the initial handshake. The server does not use OAuth, so your client should not prompt you to sign in through a browser — if it does, the API key header is not reaching us.

---

Tools

Call tools/list for the authoritative list. These are the tools available today:

Tool	Arguments	What you get	Debits balance
get_case	id (required), max_chars	Court, date, judges, parties, advocates, cited acts, citations, and the judgment text.	Yes
get_act	id (required), max_chars	Act name, section number and title, state, year, and the section text.	Yes
get_lawyer	slug or id	Published advocate profile: practice areas, courts, bar council, experience, languages.	Yes
search_cases	query (required), page	Ranked judgment results for a keyword or phrase, with citations and a matching excerpt.	Yes
search_acts	query (required), page	Ranked bare act sections for a keyword or phrase.	Yes
search_lawyers	query (required), page	Published advocate profiles matching a name or practice area.	Yes

Search tools were added after the initial release. If your client does not show them, reconnect the server so it refreshes its tool list.

---

Billing

• Rs.0.50 per billable tool call, taken from the same prepaid balance as the REST API.
• initialize, tools/list and ping are free — you are only charged for tool calls that return data.
• A call that finds nothing, or that fails validation, is not charged.
• Your remaining balance is returned in the _meta.amount_remaining field of each billable result.
• Retries are charged again, exactly like the REST API. If your agent loops, it spends credit.
• The Rs.500 signup credit is shared with the REST API — it is one balance, not one per surface.

See pricing or recharge your account.

---

Limits

Worth knowing before you build:

• Text is truncated. Judgment and act text is capped (default 32,000 characters, maximum 80,000) so it fits in a model's context. The response sets truncated when it cut the text; raise max_chars to get more. For a complete, untruncated document, use the REST .json API.
• Advocate contact details are not available. get_lawyer never returns a phone number or email address.
• No OAuth and no browser session. Authentication is an API key header only. The endpoint sets no cookies and has no login flow.
• Rate limited. Tool calls are capped per key and per IP. Over the limit you get HTTP 429 with a Retry-After header — back off and retry rather than hammering.
• One request per call. JSON-RPC batching is not supported, so that billing stays one charge per tool call.

---

Support

Questions, or something not behaving as documented? Email [email protected] and include the tool name and roughly when the call happened.