GlassTape Policy Builder
Converts natural language security requirements into validated Cerbos YAML policies with automated testing and red-team analysis, enabling AI governance with zero-trust guardrails for tool calls, data access, and compliance frameworks.
README
🧩 GlassTape Policy Builder MCP Server
Transform natural language into production-ready AI governance policies.
GlassTape Policy Builder is an open-source MCP server that converts natural-language security requirements into Cerbos YAML policies with automated validation, testing, and red-teaming.
It enables security and engineering teams to integrate AI agents and applications with policy-as-code frameworks—bringing zero-trust guardrails to tool-call interception, data access, and model workflows.
🚀 Features
- ⚙️ Natural-Language to Policy – Generate Cerbos policies from plain English using Claude or AWS Q
- 🧠 Automated Validation – Uses the Cerbos CLI (
cerbos compile,cerbos test) for syntax and logic checks - 🧪 Red-Team Analysis – 6-point security analysis with automatic improvement suggestions
- 🧩 MCP Integration – Works natively in IDEs like Cursor, Zed, and Claude Desktop
- 🔒 Air-Gapped Operation – Local-first design with no external dependencies
- 🏷️ Topic-Based Governance – 40+ content topics with safety categorization
- 🧾 Compliance Templates – Built-in templates for SOX, HIPAA, PCI-DSS, and EU AI Act
🚀 Quick Start
1. Prerequisites
Install Cerbos CLI (required for policy validation):
# 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 installation
cerbos --version
2. Install from Source
# 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 .
# With optional LLM support (for server-side natural language parsing)
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]"
3. Configure Your MCP Client
Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"glasstape-policy-builder": {
"command": "glasstape-policy-builder-mcp"
}
}
}
Cursor/Zed: Add similar configuration in your IDE's MCP settings.
Optional: Server-side LLM (for natural language processing):
{
"mcpServers": {
"glasstape-policy-builder": {
"command": "glasstape-policy-builder-mcp",
"env": {
"LLM_PROVIDER": "anthropic",
"ANTHROPIC_API_KEY": "sk-ant-your-key"
}
}
}
}
4. Usage Examples
Generate a Policy (in Claude Desktop or MCP-enabled IDE):
Create a payment policy for AI agents:
- Allow payments up to $50
- Block sanctioned entities
- Limit to 5 transactions per 5 minutes
List Available Templates:
list_templates
Validate a Policy:
validate_policy with policy_yaml: "<your-cerbos-yaml>"
5. Troubleshooting
Cerbos CLI not found:
- Ensure Cerbos CLI is installed and in your PATH
- Run
cerbos --versionto verify installation (note:--versionnotversion)
MCP server not connecting:
- Check your MCP client configuration
- Restart your IDE after configuration changes
- Verify the command path is correct:
which glasstape-policy-builder-mcp
Installation fails with "Unable to determine which files to ship":
- This is a known hatch build issue - ensure you're in the correct directory
- The pyproject.toml should include
[tool.hatch.build.targets.wheel]configuration
Import errors with MCP:
- Ensure you have the correct MCP imports:
from mcp.server import Server - Try reinstalling:
pip install -e . --force-reinstall
Policy validation fails:
- Check YAML syntax in generated policy
- Ensure Cerbos CLI is working:
cerbos compile --help - Review error messages for specific issues
Command not found after installation:
- Ensure you have Python 3.10 or higher
- Check that the entry point is correctly configured in pyproject.toml
🦭 Available Tools
When connected via MCP, you can use these tools in Claude or your IDE:
| Tool | What it does |
|---|---|
generate_policy |
Transform natural language → 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) |
Example workflow:
1. "Generate a payment policy for AI agents with $50 limit..."
→ Claude calls generate_policy
2. "Show me available financial templates"
→ Claude calls list_templates
3. "Test this policy with the test suite"
→ Claude calls test_policy
4. "Analyze this policy for security issues"
→ Claude calls suggest_improvements
5. "Validate the policy syntax"
→ Claude calls validate_policy
🧪 Example Output
Input:
"Allow AI agents to execute payments up to $50. Block sanctioned entities.
Limit cumulative hourly amount to $50. Maximum 5 transactions per 5 minutes."
Generated Policy with Topic Governance:
# policies/payment_policy.yaml
apiVersion: api.cerbos.dev/v1
resourcePolicy:
version: "1.0.0"
resource: "payment"
rules:
- actions: ["execute"]
effect: EFFECT_ALLOW
condition:
match:
expr: >
request.resource.attr.amount > 0 &&
request.resource.attr.amount <= 50 &&
!(request.resource.attr.recipient in request.resource.attr.sanctioned_entities) &&
(request.resource.attr.cumulative_amount_last_hour + request.resource.attr.amount) <= 50 &&
request.resource.attr.agent_txn_count_5m < 5 &&
has(request.resource.attr.topics) &&
"payment" in request.resource.attr.topics &&
!("adult" in request.resource.attr.topics)
- actions: ["*"]
effect: EFFECT_DENY
Plus:
- ✅ Topic-based governance (payment, pii detection)
- ✅ Safety categorization (G/PG/PG_13/R/adult_content)
- ✅ 15+ automated test cases
- ✅ Validated by
cerbos compile - ✅ 6-point security analysis
- ✅ Ready-to-deploy bundle
📋 Complete Examples
| Category | Example | Description |
|---|---|---|
| Finance | payment_policy.md | Payment execution with limits |
| Healthcare | phi_access_policy.md | HIPAA-compliant PHI access |
| AI Safety | ai_model_invocation_policy.md | Model invocation with guardrails |
| Data Access | pii_export_policy.md | GDPR-compliant PII export control |
| System | admin_access_policy.md | Admin access with MFA |
See examples/README.md for complete examples.
🧱 Architecture
flowchart TD
A["Natural-language policy request"] --> B["GlassTape MCP Server"]
B --> C["Intermediate Canonical Policy - JSON"]
C --> D["Cerbos YAML policy generation"]
D --> E["Cerbos CLI validation + testing"]
E --> F["Ready-to-deploy policy bundle"]
Key Innovation: ICP (Intermediate Canonical Policy) serves as a language-agnostic intermediate representation, enabling deterministic generation, policy portability, and formal verification.
🧪 Development
# Clone and setup
git clone https://github.com/glasstape/glasstape-policy-builder-mcp.git
cd glasstape-policy-builder-mcp
pip install -e ".[dev]"
# Run tests
pytest
# Format code
black src/ tests/
🤝 Contributing
We welcome contributions! See CONTRIBUTING.md for guidelines.
Quick Links:
💪 License
Released under the Apache 2.0 License. © 2025 GlassTape, Inc.
💡 Links
Built with ❤️ by GlassTape — Making AI agents secure by default.
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.