baymax-legal 1.0.0

Extraction-based legal research synthesis tool for Claude Desktop

Baymax - Legal Research Synthesis Tool

Zero hallucination. Pure extraction.

Baymax is a powerful MCP server for Claude Desktop that provides extraction-based legal research tools. Unlike AI that generates or summarizes, Baymax extracts actual text from cases - holdings, rules, facts, and procedural posture - with full citations.

Quick Start (5 minutes)

Step 1: Install Python Dependencies

cd "path/to/baymax"
pip install -r requirements.txt

Step 2: Configure Claude Desktop

Add Baymax to your Claude Desktop config file:

Windows: %APPDATA%\Claude\claude_desktop_config.json Mac: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "baymax": {
      "command": "python",
      "args": ["C:/path/to/baymax/citation_mcp_text_based.py"]
    }
  }
}

Replace the path with your actual installation location.

Step 3: Restart Claude Desktop

Close and reopen Claude Desktop. You should see "baymax" in your MCP servers.

Step 4: Start Using

Just ask Claude:

• "Brief this case for me" (paste case text)
• "Extract the holding from Brady v. Maryland"
• "Synthesize the rules across these three cases"
• "Generate a parenthetical for this case"

---

Features

Core Extraction Tools

Tool	Description
extract_citation	Parse case names and citations from text
extract_rules	Extract holdings, standards, elements, and factors
extract_case_facts	Extract procedural posture, key facts, and outcome
generate_parenthetical	Create properly formatted case parentheticals
build_citation_string	Format citations with signals and pin cites

Workflow Tools (One-Click Analysis)

Tool	Description
brief_this_case	Complete case briefing - facts, rules, holding, parenthetical
synthesize_rules	Multi-case rule synthesis with hierarchy
validate_brief	Check brief for citation errors and unsupported assertions

Research Tools

Tool	Description
find_similar_holdings	Match arguments to case holdings
suggest_signals	Recommend citation signals based on context
analyze_argument_strength	Evaluate supporting authority strength

Deposition Tools

Tool	Description
ingest_deposition	Parse transcript into searchable Q&A entries
search_deposition	Find testimony by keyword, topic, or citation
validate_facts_against_record	Check statement of facts against depositions

---

Installation Options

Basic (MCP Server Only)

pip install mcp

Recommended (With PDF Support)

pip install mcp PyPDF2 pdfplumber python-docx

Full Install (All Features)

pip install -r requirements.txt

Optional: Web UI

Baymax includes a web interface for drag-and-drop case analysis:

pip install flask
python web_app.py
# Open http://127.0.0.1:5000

---

Usage Examples

Brief a Case

User: Brief this case for me:
[paste case text]

Claude: Uses brief_this_case tool...

CASE BRIEF: Brady v. Maryland
=============================
CITATION: 373 U.S. 83 (1963)

PROCEDURAL POSTURE:
Certiorari to the Court of Appeals of Maryland

KEY FACTS:
- Petitioner and companion were convicted of murder
- Companion admitted the actual killing
- Prosecution withheld companion's confession

HOLDING:
The suppression by the prosecution of evidence favorable
to an accused upon request violates due process where the
evidence is material either to guilt or to punishment.

PARENTHETICAL:
(holding that prosecution's suppression of evidence
favorable to accused violates due process when material
to guilt or punishment)

Synthesize Rules Across Cases

User: Synthesize the qualified immunity rules from these cases:
[paste 3 case texts]

Claude: Uses synthesize_rules tool...

RULE SYNTHESIS: Qualified Immunity
==================================
MASTER RULE:
Government officials are entitled to qualified immunity
unless their conduct violated clearly established law.

ELEMENTS:
1. Whether the facts show a constitutional violation
2. Whether the right was clearly established at the time
3. Whether a reasonable official would have known

SOURCES: Harlow v. Fitzgerald, Saucier v. Katz, Pearson v. Callahan

Validate a Brief

User: Check this brief section for problems:
[paste brief section]

Claude: Uses validate_brief tool...

VALIDATION REPORT
=================
CITATIONS: 5 found, 4 valid

ISSUES:
[!] "Smith v. Jones, 123 F.3d 456" - citation format error
[!] Assertion on page 3 lacks supporting authority
[!] Signal "see" may be incorrect - consider "see also"

SUGGESTIONS:
- Add pin cite for block quote from Johnson v. State
- Consider distinguishing adverse authority on element 2

---

Requirements

• Python 3.9+
• Claude Desktop with MCP support
• Optional: Tesseract OCR (for scanned PDFs)
• Optional: Poppler (for PDF-to-image conversion)

File Structure

baymax/
├── citation_mcp_text_based.py   # MCP server (main entry point)
├── brief_synthesis.py           # Parentheticals, synthesis
├── enhanced_rules.py            # Rule/holding extraction
├── extraction_validation.py     # Validation tools
├── workflows.py                 # Workflow orchestration
├── deposition_tools.py          # Deposition processing
├── web_app.py                   # Optional web UI
├── templates/
│   └── index.html               # Web UI template
├── requirements.txt             # Python dependencies
├── pyproject.toml               # Package configuration
└── README.md                    # This file

Troubleshooting

"MCP server not connecting"

1. Check the path in claude_desktop_config.json is correct
2. Ensure Python is in your PATH
3. Try running python citation_mcp_text_based.py directly to see errors

"Module not found"

pip install mcp

"PDF extraction not working"

pip install PyPDF2 pdfplumber

"DOCX export not working"

pip install python-docx

---

How It Works

Baymax uses pattern-based extraction - not AI generation. Every piece of text returned is extracted directly from the source document using legal-specific regex patterns:

1. Holdings: Patterns like "We hold that...", "The court held..."
2. Rules: "Under [standard], a party must..."
3. Facts: Procedural posture, party actions, key events
4. Elements: Numbered lists, "requires that" constructions

This means:

• No hallucination - only text from your documents
• Verifiable - check every extraction against the source
• Citation-ready - formatted for immediate use

---

License

MIT License - Free for personal and professional use.

Contributing

Issues and PRs welcome at the project repository.

---

Baymax - Extraction-based legal research synthesis for Claude Desktop.