CSL Core

Provides deterministic safety enforcement for AI workflows by loading CSL policies, verifying with Z3, and enforcing at runtime.

python

2

GitHub Stars

python

Language

3 months ago

First Indexed

2 months ago

Catalog Refreshed

Documentation & install

Readme and setup notes from the catalogue, plus a client-ready config you can copy for your MCP host.

Installation

Add the following to your MCP client configuration file.

Configuration

View docs

{
  "mcpServers": {
    "chimera-protocol-csl-core": {
      "command": "uv",
      "args": [
        "run",
        "--with",
        "csl-core[mcp]",
        "csl-core-mcp"
      ]
    }
  }
}

CSL-Core MCP Server provides a deterministic, outside-the-model safety enforcement layer for AI systems. You install the MCP integration, load CSL policy files, and run a local service that compiles and verifies policies with Z3, then enforces them in real time. This enables you to guarantee policy compliance while keeping the rules separate from the AI model, ensuring consistent, auditable behavior.

How to use

Set up CSL-Core MCP to protect your AI workflows by loading CSL policies and routing tool calls through the policy guard. You will verify, simulate, and enforce policies via the MCP server from your client application, enabling deterministic decisions and zero-on-model rule leakage.

How to install

Prerequisites: you need Python and the pip package manager installed on your system.

pip install "csl-core[mcp]"

Next, configure the MCP server in your environment so your client can connect securely. Add the following MCP server entry to your Claude Desktop configuration to enable the CSL-Core MCP integration.

{
  "mcpServers": {
    "csl-core": {
      "command": "uv",
      "args": ["run", "--with", "csl-core[mcp]", "csl-core-mcp"]
    }
  }
}

To start the MCP process directly (for development or non-Claude usage), run the exact command described above in your shell: "uv run --with csl-core[mcp] csl-core-mcp".

Additional notes

The MCP server executes policy verification at compile-time and runtime, offering deterministic enforcement with sub-millisecond latency. It uses Z3 for formal verification and supports integration patterns like LangChain, CLI tools, and automated simulations.

Configuration and usage highlights

Policies are written in CSL files and verified with Z3 before they can be enforced at runtime.
The MCP server supports interactive verification, scenario simulation, and policy explanation endpoints.

Available tools

verify_policy

Z3 formal verification to catch contradictions at compile time and ensure policy consistency.

simulate_policy

Test policies against JSON inputs to observe whether actions are ALLOWED or BLOCKED.

explain_policy

Provide a human-readable summary of a CSL policy, including constraints and RBAC rules.

scaffold_policy

Generate a CSL template from plain-English descriptions to jumpstart policy creation.