supreme-court-drafting-mcpb

<div align="center"> <img src="docs/banner.png" width="820"/> **A drafting connector for the Supreme Court of India.** Visit the live site: [wolfgangrush.github.io](https://wolfgangrush.github.io)     </div> # wolfgang_rush — Supreme Court Drafting **MCPB Desktop Extension** for drafting pleadings before the Supreme Court of India. Designed for Indian advocates using **Claude Desktop App**. Local-execution. Zero data collection. > *Also available as a Claude Code Plugin (for developers using the Claude Code CLI):* > *[github.com/Wolfgangrush/supreme-court-drafting-litigation](https://github.com/Wolfgangrush/supreme-court-drafting-litigation)* --- ## What this connector does Exposes the wolfgang_rush six-agent SC drafting pipeline as MCP tools that Claude can orchestrate from a Claude Desktop chat. Drafts the following case types: | Case type | Statutory anchor | |---|---| | SLP (Civil) | Article 136 of the Constitution | | SLP (Criminal) | Article 136 of the Constitution | | Writ Petition (Article 32) | Article 32 of the Constitution | | Transfer Petition | Section 25 CPC / Section 406 CrPC | | Review Petition | Article 137 + Order XLVII SC Rules 2013 | | Curative Petition | Rupa Hurra (2002) 4 SCC 388 framework | Output is rendered as filing-grade `.docx` via pandoc using the bundled SC reference.docx (A4, Times New Roman 14, 1.5 line spacing, 4 cm left margin). --- ## Install 1. Open **Claude Desktop App** 2. **Settings → Extensions → Install Extension** 3. Select `wolfgang-supreme-court-drafting.mcpb` 4. Enable the extension 5. Start a new chat ## System requirements - macOS, Windows, or Linux with Claude Desktop App ≥ 0.10.0 - Python ≥ 3.10 (Claude Desktop App bundles uv runtime automatically) - **`pandoc`** for `.docx` output (`brew install pandoc` on macOS · `apt-get install pandoc` on Linux · download from pandoc.org on Windows) - **`pdftotext`** for PDF reading (`brew install poppler` on macOS · `apt-get install poppler-utils` on Linux) ## Usage In a Claude Desktop chat, simply describe what you want to draft. Claude will discover and call the connector's tools as needed. Example prompts: - *"Draft an SLP-civil under Article 136 against the Bombay HC dismissal dated 2026-04-10, with grounds based on perversity of finding on issue 3."* - *"Draft a writ petition under Article 32 challenging Section X of [ACT-Y] on Article 14 and 21 grounds, with prayer for interim stay."* - *"Draft a curative petition within the Rupa Hurra limbs — gross miscarriage of justice plus abuse of process."* Claude will: 1. Call `list_case_types()` to discover available case types 2. Call `get_case_type_format(case_type)` to retrieve the drafting template 3. Call `get_pleading_base()` to retrieve the universal SC pleading structure 4. Optionally call `read_case_folder(path)` if you point it at a case-files folder 5. Draft the pleading 6. Optionally call `save_draft_as_docx(...)` to render the final `.docx` ## Tools | Tool | Purpose | |---|---| | `list_case_types` | Discover available SC case types | | `get_case_type_format` | Retrieve the drafting template for a specific case type | | `get_agent_instructions` | Retrieve instructions for a stage in the six-agent pipeline | | `get_pleading_base` | Retrieve the universal SC pleading skeleton + Registry formatting | | `read_case_folder` | Read files in a case folder for fact extraction | | `save_draft_as_docx` | Render markdown as filing-grade .docx | ## Privacy This connector collects **zero** user data. All processing happens on the user's machine. The publisher (wolfgang_rush) never receives any user data. The connector applies a three-layer privacy firewall recommended for all case-folder work: - **L1 substitution** — real party names, case numbers, financial figures replaced with neutral placeholders before any text is sent to Claude - **L2 LLM-blind** — Claude sees only placeholders, never the real privileged data - **L3 re-substitution** — placeholders mapped back to real values at the `.docx` render stage on the user's machine Read the canonical privacy policy at **<https://wolfgangrush.github.io/privacy/>**. ## Confidentiality and professional privilege Case-folder contents passed to this connector may include material attracting attorney-client privilege under **Section 132 of the Bharatiya Sakshya Adhiniyam 2023** / **Section 126 of the Indian Evidence Act 1872**. Decisions about what material to supply remain the user's professional responsibility consistent with applicable Bar Council of India rules and client-engagement terms. ## Architecture · how the six agents work This connector runs a strict six-agent pipeline locally on your machine: | Agent | What it does | Output | |---|---|---| | **Reader** | Reads every input document. Extracts facts with per-document audit log. **Applies the pseudonymisation firewall** (see below). Halts if a required statute PDF is missing. | `case-facts.md` | | **Format** | Loads the case-type-specific skill + bench/state/forum-config + pleading base. Maps Reader's facts into the format placeholders. | `format-shell.md` | | **Drafter** | Writes the first complete draft — Cause Title, Statutory Opening, Synopsis, Statement of Facts, Grounds, Prayer, Verification, Counsel Block, Index, Annexure List. | `draft-v1.md` + `draft-v1.docx` | | **Verifier** | Anti-hallucination firewall. Compares draft-v1 against case-facts.md fact-by-fact. Flags hallucinated dates, fabricated citations, unsupported assertions, orphan annexure markers, missing factual basis. | `verification-report.md` | | **Refiner** | Applies every Verifier flag. Polishes language to formal Indian pleading register. Enforces Registry formatting. Strips AI-style markers. | `draft-v2.md` + `draft-v2.docx` | | **Overseer** | Reads draft-v2 with opposing-counsel lens. Finds weak prayers, contradictory facts, attackable defects, missing limbs of argument. Suggests hardening. | `opposing-notes.md` + `final-draft.docx` | The pipeline is **forced by the connector itself** — the `get_agent_instructions()` tool is the mandatory first call when you ask for a draft and returns an 11-step orchestration script that names every agent's tool call. The Drafter cannot legitimately produce final output without the Reader having saved `case-facts.md` first. The `save_artifact` tool's allow-list rejects standalone python-docx or JavaScript generator scripts. → **Full pipeline architecture: [wolfgangrush.github.io/mcpb/agents/](https://wolfgangrush.github.io/mcpb/agents/)** ## 🔒 Pseudonymisation gateway — what gets substituted The Reader agent applies a privacy firewall **before any downstream agent sees the facts**. The following are substituted with structural placeholders: - **Party identifiers** — Petitioner / Respondent / Plaintiff / Defendant / Accused / Complainant / Witness names → `[Petitioner-A]`, `[Respondent-B]`, `[Witness-A]` - **Addresses** — Full residential / business addresses → `[Address-Placeholder]` - **Government identifiers** — PAN, Aadhaar, TAN, DIN, GSTIN → `[PAN-Placeholder]`, `[Aadhaar-Placeholder]`, etc. - **Case numbers** — FIR / CR / Crime / SLP / Diary / CC / SC / RCS / lower-court case numbers → `[Crime-No-Placeholder]`, `[SLP-No-Placeholder]`, `[Lower-Court-Case-No-Placeholder]` - **Financial figures** — Amounts in dispute, compensation, tax assessed → `[Amount-Placeholder]` - **Statutory notice references** — Section 106 TPA notice dates, statutory demand-notice dates → `[Notice-Date-Placeholder]` The Drafter, Verifier, Refiner, and Overseer agents process **placeholders only**. At the final `save_draft_as_docx` step, the placeholders are re-substituted with the real values **on your local machine**. The LLM never sees the re-substituted output. This is the connector's contribution to your **Section 8(5) DPDP Act 2023** safeguard. → **Full pseudonymisation mechanism: [wolfgangrush.github.io/mcpb/agents/#pseudonymisation-gateway-what-gets-substituted](https://wolfgangrush.github.io/mcpb/agents/#pseudonymisation-gateway--what-gets-substituted)** ## ⚖️ DPDP Act 2023 — what this means for you **Publisher position.** wolfgang_rush, in its capacity as software publisher, is **neither a Data Fiduciary nor a Data Processor** under the DPDP Act 2023 in respect of this connector. The connector runs entirely on your machine. There is no wolfgang_rush server, no telemetry, no API endpoint that the publisher controls. Section 2(i) requires "determining purpose and means of processing" — wolfgang_rush determines neither. **User position.** You — the advocate using this connector — are the **Data Fiduciary** for your own client's personal data. This was true before installing the connector and remains true after. Your obligations under Sections 4, 5, 6, 8, 9, 11, 13 of the DPDP Act 2023 continue independent of this connector. **What the connector helps with.** The pseudonymisation gateway is an architectural safeguard within the meaning of Section 8(5) (reasonable security safeguards). Local-only processing supports your minimisation posture (Section 8). The Reader's per-document audit log supports Section 8(8) (data accuracy). The Section 17(2)(a) exemption ("personal data processed for the purposes of any legal proceeding") substantially covers most everyday advocate processing. **What the connector does not do.** It does not, by itself, satisfy any DPDP notice / consent / grievance-redressal obligation. Those remain yours to operationalise. It does not cover Anthropic's position as the LLM operator — that is governed by Anthropic's own terms. → **Full DPDP applicability analysis: [wolfgangrush.github.io/mcpb/dpdp/](https://wolfgangrush.github.io/mcpb/dpdp/)** ## Multilingual install guides [हिन्दी](https://wolfgangrush.github.io/mcpb/hi/) · [मराठी](https://wolfgangrush.github.io/mcpb/mr/) · [தமிழ்](https://wolfgangrush.github.io/mcpb/ta/) · [తెలుగు](https://wolfgangrush.github.io/mcpb/te/) · [বাংলা](https://wolfgangrush.github.io/mcpb/bn/) · [ગુજરાતી](https://wolfgangrush.github.io/mcpb/gu/) · [ಕನ್ನಡ](https://wolfgangrush.github.io/mcpb/kn/) · [ਪੰਜਾਬੀ](https://wolfgangrush.github.io/mcpb/pa/) · [മലയാളം](https://wolfgangrush.github.io/mcpb/ml/) · [اردو](https://wolfgangrush.github.io/mcpb/ur/) ## ⚠️ AI verification disclaimer · 🔒 Pseudonymisation procedure > **⚠️ AI can make mistakes — please verify the information before filing.** > Every draft produced by this connector is a STARTING POINT. The Verifier > agent runs an anti-hallucination firewall and the Overseer agent runs an > opposing-counsel review, but neither replaces an advocate's independent > verification of statutory references, citation accuracy, factual fidelity, > and Registry-formatting compliance with the user's High Court / forum. > The advocate filing the pleading remains responsible for the contents. > > **🔒 Protected by pseudonymisation procedure.** The Reader agent applies a > domain-specific privacy firewall as the first step of the pipeline — party > names, addresses, identifying numbers (FIR / CR / Crime / Suit / Diary / > SLP / lower-court case numbers), PAN / Aadhaar references, financial > figures, witness names, and statutory-notice references are substituted > with structural placeholders BEFORE any downstream agent sees the facts. > The Drafter, Verifier, Refiner, and Overseer agents process placeholders > only. Real values are re-substituted at the final docx render step on the > user's local machine. No real identifying data leaves the case folder. ## License MIT. See LICENSE. ## Publisher **wolfgang_rush** — independent open-source legal-tech publisher. Contact: wolfgangrush@gmail.com ## Source <https://github.com/Wolfgangrush/supreme-court-drafting-mcpb> ## Sample cases See `SAMPLE-CASES/` for three anonymised fact patterns the reviewer can use to invoke the tools.