ai-control-framework

The AI Control Framework MCP server provides a disciplined, production-focused evaluation workflow for AI pilots. It exposes mechanisms to enforce contract integrity, mock timeout policies, and comprehensive deployment readiness scoring (DRS) to help teams ship stable AI-powered features. When integrated with an MCP runner (for example, Claude Code workflows), you can run the framework’s checks across your session artifacts, generate a deployability score, and receive actionable guidance on what to fix before production. The server leverages a set of scripts and prompts designed to verify contract adherence, detect lingering mocks, and calculate a multi-component DRS score that emphasizes deploy readiness. Use the provided tools to assess, validate, and iterate on AI-driven sessions until you reach a high confidence level for deployment.

Prerequisites:

• Node.js and npm installed on your system
• Git available to clone repositories

Steps:

1. Install the MCP server package (example uses npx):

   • npx -y ai-control-framework

2. If you prefer to install locally, clone the repository and install dependencies:

   • git clone https://github.com/sgharlow/ai-control-framework.git
   • cd ai-control-framework
   • npm install

3. Start or register the MCP server as described by the package’s usage (the MCP integration is designed to be invoked via the npx command or through your MCP runner). If starting locally, you may run the server entry point (adjust according to the package’s actual entry script):

   • node path/to/server.js

4. Validate the server is reachable by running a quick health check or the included quickstart commands as documented in the project README.

Note: The exact entry points may vary depending on how the MCP server is published; consult the repository’s usage section or package.json for the precise start command if you clone the repo directly.

Tips and common issues:

• Ensure Node.js is a compatible version with the package requirements.
• If using npx, you may need network access to fetch the package on first run.
• Environment variables: DRS_TARGET is the deployability threshold for shipping; MOCK_TIMEOUT defines the window (in minutes) for mock usage before requiring real service calls. Adjust these in the mcp_config env as needed for your environment.
• If you encounter contract violations or hash mismatches during checks, verify your OpenAPI/contract definitions and re-run the contract-freezing checks as described by the framework.
• When integrating with your CI/CD, ensure the MCP server invocation aligns with your workflow’s authentication and permissions model (e.g., GitHub Actions, self-hosted runners).
• The MCP server is designed to work with Claude Code integrations; if you’re using another agent, confirm compatibility and required prompts/files.
• If you need to override the default DRS components, you can tune the DRS calculation scripts or scoring rubrics as per your deployment requirements.