legalis-mcp 0.2.0

MCP server for Legalis — Brazilian legal document AI for attorneys

Legalis MCP Server

MCP (Model Context Protocol) server for Legalis — Brazilian legal document AI for attorneys.

Connect Claude (or any MCP-compatible client) to your Legalis account and interact with cases, generate legal documents, search legislation, and more using natural language.

---

Quick start

1. Install

pip install legalis-mcp
# or with uv:
uv tool install legalis-mcp

2. Authenticate

legalis-mcp auth

This opens your browser. Log in to Legalis normally — tokens are saved to ~/.legalis/credentials.json (owner-readable only).

3. Configure Claude

Add to your Claude configuration (~/.claude/claude_desktop_config.json for Claude Desktop, or claude_code_config.json for Claude Code):

{
  "mcpServers": {
    "legalis": {
      "command": "legalis-mcp",
      "args": ["serve"]
    }
  }
}

4. Use

In Claude, try:

• "Liste meus casos no Legalis"
• "Crie um caso de rescisão indireta para João Silva"
• "Gere a seção de fundamentos jurídicos da peça X do caso Y"
• "Pesquise legislação trabalhista sobre justa causa"
• "Quantas gerações ainda tenho disponíveis este mês?"

---

Available tools

Tool	Description
list_cases	List cases (paginated, filterable by status)
get_case	Get case details by ID
create_case	Create a new case
update_case	Update case fields
search_cases	Full-text search across cases
list_legal_documents	List peças processuais for a case
get_legal_document	Get a document with all sections
create_legal_document	Create a new legal document
generate_section	AI-generate a document section (fatos, fundamentos, pedidos…)
get_case_context	Get case context (facts, parties, evidence)
run_interview	AI-guided case intake interview
search_legislation	Search Brazilian legislation (LexML)
get_my_usage	Check AI quota and billing period usage

---

CLI commands

legalis-mcp auth      # Authenticate via browser
legalis-mcp serve     # Start MCP server (stdio)
legalis-mcp status    # Verify authentication
legalis-mcp logout    # Remove stored credentials

---

Development

git clone <repo>
cd mcp
pip install -e ".[test]"

# Run unit tests (no external services needed)
pytest tests/unit/ -v

# Run integration tests (requires Docker)
cd ../integration && docker-compose up -d
cd ../mcp
LEGALIS_API_URL=http://localhost:8000 \
LEGALIS_ACCESS_TOKEN=<token> \
LEGALIS_REFRESH_TOKEN=<refresh> \
pytest tests/integration/ -v

Environment variables

Variable	Default	Description
LEGALIS_API_URL	https://hai-production-612f.up.railway.app	Backend URL
LEGALIS_ACCESS_TOKEN	—	Pre-issued token (CI/test only)
LEGALIS_REFRESH_TOKEN	—	Refresh token (CI/test only)
LEGALIS_CREDENTIALS_FILE	~/.legalis/credentials.json	Credentials file path
LEGALIS_OAUTH_PORT	3742	Local OAuth callback port

---

Security

• Credentials are stored in ~/.legalis/credentials.json with 600 permissions (owner read/write only).
• The OAuth redirect is restricted to http://localhost — open-redirect attacks are blocked at the backend.
• Tokens are never stored in the MCP configuration file.
• All API calls go through the existing Legalis backend — RBAC, rate limiting, and quotas apply normally.

---

Publishing

# PyPI
pip install build twine
python -m build
twine upload dist/*

# Then register at:
# - smithery.ai (MCP marketplace)
# - mcp.so (community directory)
# - github.com/modelcontextprotocol/servers (official list)