Cerebra Legal MCP Server

An enterprise-grade MCP server for legal reasoning and analysis based on the "think" tool concept from Anthropic's engineering blog.

Overview

Cerebra Legal provides three powerful tools for legal reasoning and analysis:

legal_think - A structured legal reasoning tool that helps analyze complex legal issues with domain-specific guidance and templates.
legal_ask_followup_question - A specialized tool for asking follow-up questions in legal contexts with domain-specific options.
legal_attempt_completion - A tool for presenting legal analysis results with proper structure and citation formatting.

The server automatically detects legal domains (ANSC contestation, consumer protection, contract analysis) and provides domain-specific guidance, templates, and feedback.

Features

Domain Detection: Automatically identifies the legal domain of the analysis
Domain-Specific Guidance: Provides tailored guidance for different legal domains
Structured Templates: Offers domain-specific templates for legal analysis
Citation Formatting: Properly formats legal citations
Thought Quality Analysis: Provides feedback on legal reasoning quality
Revision Support: Allows for revising previous thoughts

Installation

# Clone the repository
git clone https://github.com/yoda-digital/mcp-cerebra-legal-server.git
cd mcp-cerebra-legal-server

# Install dependencies
npm install

# Build the project
npm run build

Usage

Running the Server

npm start

Testing the Server

The repository includes a test client that demonstrates how to interact with the server:

# Make the test client executable
chmod +x test-client.js

# Run the test client
./test-client.js

The test client will:

Start the server
Send a tools/list request to get available tools
Send a legal_think request with a sample thought
Display the server's responses

Adding to Claude

To add the server to Claude, update your MCP settings file with the following configuration:

For VSCode Extension

Edit the file at ~/.config/Code/User/globalStorage/rooveterinaryinc.roo-cline/settings/cline_mcp_settings.json:

{
  "mcpServers": {
    "cerebra-legal": {
      "command": "node",
      "args": ["/path/to/mcp-cerebra-legal-server/build/index.js"],
      "disabled": false,
      "alwaysAllow": []
    }
  }
}

For Claude Desktop App

Edit the file at ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on your platform:

{
  "mcpServers": {
    "cerebra-legal": {
      "command": "node",
      "args": ["/path/to/mcp-cerebra-legal-server/build/index.js"],
      "disabled": false,
      "alwaysAllow": []
    }
  }
}

Using the Tools in Claude

Once the server is added to Claude, you can use the tools in your conversations:

System Prompt Integration

For optimal AI interaction with these legal tools, it's essential to include proper guidance in the system prompt. These tools follow the same AI → Human interaction model as standard Cline tools but with domain-specific enhancements.

The system prompt should include detailed documentation for each tool:

## legal_think
Description: A structured legal reasoning tool that helps analyze complex legal issues with domain-specific guidance and templates. This tool enables step-by-step legal analysis with feedback on reasoning quality.

Parameters:
- thought: (required) The current legal reasoning step
- thoughtNumber: (required) Current thought number in sequence
- totalThoughts: (required) Estimated total thoughts needed
- nextThoughtNeeded: (required) Whether another thought step is needed
- category: (optional) Legal domain category (auto-detected if not provided)
- references: (optional) Array of legal references
- isRevision: (optional) Whether this revises previous thinking
- revisesThoughtNumber: (optional) Which thought is being reconsidered
- requestGuidance: (optional) Whether to request domain-specific guidance
- requestTemplate: (optional) Whether to request a domain-specific template

Usage:
<legal_think>
<thought>Your legal reasoning step here</thought>
<thoughtNumber>Current thought number</thoughtNumber>
<totalThoughts>Estimated total thoughts</totalThoughts>
<nextThoughtNeeded>true or false</nextThoughtNeeded>
<category>Legal domain (optional)</category>
<references>Array of references (optional)</references>
<isRevision>true or false (optional)</isRevision>
<revisesThoughtNumber>Thought number being revised (optional)</revisesThoughtNumber>
<requestGuidance>true or false (optional)</requestGuidance>
<requestTemplate>true or false (optional)</requestTemplate>
</legal_think>

## legal_ask_followup_question
Description: Ask the user a legal domain-specific question to gather additional information needed to complete the task. This tool enhances the standard ask_followup_question with legal domain detection, terminology formatting, and domain-specific suggested options.

Parameters:
- question: (required) The question to ask the user. This will be automatically enhanced with appropriate legal terminology.
- options: (optional) An array of 2-5 options for the user to choose from. If not provided, domain-specific options will be automatically suggested.
- context: (optional) Additional context to help with domain detection and question formatting.

Usage:
<legal_ask_followup_question>
<question>Your question here</question>
<options>
Array of options here (optional), e.g. ["Option 1", "Option 2", "Option 3"]
</options>
<context>Additional context to help with domain detection (optional)</context>
</legal_ask_followup_question>

## legal_attempt_completion
Description: Present the result of your work to the user with proper legal structure and formatting. This tool enhances the standard attempt_completion with legal domain detection, document structuring, and citation formatting.

Parameters:
- result: (required) The result of the task. This will be automatically formatted with proper legal structure.
- command: (optional) A CLI command to execute to show a live demo of the result to the user.
- context: (optional) Additional context to help with domain detection and result formatting.

Usage:
<legal_attempt_completion>
<result>
Your final result description here
</result>
<command>Command to demonstrate result (optional)</command>
<context>Additional context to help with domain detection (optional)</context>
</legal_attempt_completion>

This guidance ensures the AI understands:

These are specialized versions of standard tools
They maintain the same AI → Human interaction flow
They have additional capabilities and parameters
How to properly format the tool calls

Without this guidance, the AI might not fully leverage the domain-specific capabilities built into these tools.

1. Using legal_think

The legal_think tool helps you analyze complex legal issues with structured thinking:

I need to analyze an ANSC contestation where a claimant argues that technical specifications in a tender were too restrictive.

Claude will use the legal_think tool to:

Detect the legal domain (ANSC contestation)
Provide domain-specific guidance
Offer a structured template for analysis
Give feedback on the quality of legal reasoning
Support revision of previous thoughts

2. Using legal_ask_followup_question

When Claude needs more information to complete a legal analysis:

What specific provisions of the technical specifications are being challenged?

Claude will use the legal_ask_followup_question tool to:

Format the question with appropriate legal terminology
Provide domain-specific options for the user to choose from
Detect the legal domain for context-aware questioning

3. Using legal_attempt_completion

When Claude is ready to present the final legal analysis:

Based on my analysis, the technical specifications requiring "minimum 5 years experience" appear disproportionate and likely violate Article 33(2) of Law 131/2015 on public procurement.

Claude will use the legal_attempt_completion tool to:

Format the conclusion with proper legal structure
Extract and format legal citations
Organize the analysis into clear sections
Provide a professional legal document format

Tool Input Schemas

legal_think

{
  "thought": "Analyzing ANSC contestation where claimant argues technical specifications were too restrictive.",
  "thoughtNumber": 1,
  "totalThoughts": 5,
  "nextThoughtNeeded": true,
  "category": "ansc_contestation",  // Optional, auto-detected if not provided
  "references": ["Law 131/2015", "ANSC Decision #12345"],  // Optional
  "isRevision": false,  // Optional
  "revisesThoughtNumber": null,  // Optional
  "requestGuidance": true,  // Optional
  "requestTemplate": true  // Optional
}

legal_ask_followup_question

{
  "question": "What specific provisions of the technical specifications are being challenged?",
  "options": [  // Optional, auto-generated if not provided
    "Are you challenging the experience requirements?",
    "Are you challenging the technical capacity requirements?",
    "Are you challenging the financial requirements?",
    "Are you challenging the certification requirements?"
  ],
  "context": "ANSC contestation regarding procurement of IT equipment"  // Optional
}

legal_attempt_completion

{
  "result": "Based on the analysis of ANSC contestation #12345, the technical specifications requiring 'minimum 5 years experience' appear disproportionate and likely violate Article 33(2) of Law 131/2015 on public procurement.",
  "command": null,  // Optional
  "context": "ANSC contestation analysis"  // Optional
}

Architecture

The server is built with a modular architecture:

Domain Detector: Identifies the legal domain of the analysis
Legal Knowledge Base: Provides domain-specific guidance and templates
Citation Formatter: Formats legal citations properly
Tool Implementations: Handles the logic for each tool

Development

Project Structure

mcp-cerebra-legal-server/
├── src/
│   ├── shared/           # Shared components
│   │   ├── DomainDetector.ts
│   │   ├── LegalKnowledgeBase.ts
│   │   ├── CitationFormatter.ts
│   │   └── types.ts
│   ├── tools/            # Tool implementations
│   │   ├── LegalThinkTool.ts
│   │   ├── LegalAskFollowupQuestionTool.ts
│   │   └── LegalAttemptCompletionTool.ts
│   ├── utils/            # Utilities
│   │   └── logger.ts
│   └── index.ts          # Main server entry point
├── build/                # Compiled JavaScript
├── test-client.js        # Test client
├── package.json
└── tsconfig.json

Building

npm run build

Testing

# Run the test client
./test-client.js

Repository

This project is available on GitHub at: https://github.com/yoda-digital/mcp-cerebra-legal-server

References

The "think" tool: Enabling Claude to stop and think in complex tool use situations - Anthropic Engineering Blog

License

MIT

智法大脑