GlassTape Policy Builder

GlassTape Policy Builder MCP Server turns natural-language security requirements into Cerbos YAML policies, with automated validation, testing, and red-teaming. It enables AI governance by integrating policy-as-code into your agents and applications, delivering zero-trust guardrails for tool calls, data access, and model workflows.

How to use

You connect your MCP client (such as Claude Desktop or an IDE with MCP support) to the GlassTape Policy Builder MCP Server to generate, validate, and test Cerbos policies from plain-English requirements. Use it to translate governance intents into production-ready policies, then verify them with built-in tests and a red-teaming analysis.

How to install

Prerequisites: install Python and a Python package manager. You also need the Cerbos CLI installed to validate policies.

# Install Cerbos CLI
# macOS
brew install cerbos/tap/cerbos

# Linux
curl -L https://github.com/cerbos/cerbos/releases/latest/download/cerbos_Linux_x86_64 \
  -o /usr/local/bin/cerbos && chmod +x /usr/local/bin/cerbos

# Verify
cerbos --version

Install the GlassTape Policy Builder MCP Server from source and set up the Python environment.

# Clone the repository
git clone https://github.com/glasstape/glasstape-policy-builder-mcp.git
cd glasstape-policy-builder-mcp/agent-policy-builder-mcp

# Basic installation
pip install -e .

# Optional: install LLM support (choose as needed)
pip install -e ".[anthropic]"  # Anthropic Claude
pip install -e ".[openai]"     # OpenAI GPT
pip install -e ".[llm]"        # All LLM providers

# Development installation
pip install -e ".[dev]"

Configure your MCP client

Set up your MCP client so it can call the GlassTape Policy Builder MCP Server. The following example shows configuring Claude Desktop to point to the local MCP endpoint.

{
  "mcpServers": {
    "glasstape_policy_builder": {
      "command": "glasstape-policy-builder-mcp"
    }
  }
}

Optional: server-side LLM configuration

If you want server-side natural language processing, you can provide LLM settings to the MCP client configuration.

{
  "mcpServers": {
    "glasstape_policy_builder": {
      "command": "glasstape-policy-builder-mcp",
      "env": {
        "LLM_PROVIDER": "anthropic",
        "ANTHROPIC_API_KEY": "sk-REPLACE_ME"
      }
    }
  }
}

Usage patterns

Generate a policy by describing your governance intent in natural language, then run the policy through the built-in validation and tests. You can also run a red-team analysis to obtain improvement suggestions before deployment.

Troubleshooting

If the MCP client cannot connect, verify your client configuration, restart the IDE, and ensure the command path is correct.

Tools and workflow

Use the following tools when connected via MCP to perform end-to-end policy work: generate_policy, validate_policy, test_policy, suggest_improvements, list_templates.

Security and best practices

Operate locally (air-gapped) when possible to minimize exposure. Use topic-based governance with safety categorization and leverage built-in compliance templates (SOX, HIPAA, PCI-DSS, EU AI Act) to align with regulatory requirements.

Notes

This MCP server translates natural-language requirements into Cerbos YAML policies, enabling automated validation and testing through the Cerbos CLI as part of a zero-trust policy workflow.

Available tools

generate_policy

Transform natural language into validated Cerbos YAML with topic governance

validate_policy

Check policy syntax with cerbos compile

test_policy

Run test suites against policies with cerbos compile

suggest_improvements

6-point security analysis with automatic improvement suggestions

list_templates

Browse built-in templates (finance, healthcare, AI safety)