- Published on
Conduct, open-source guardrails for LLM and MCP tool calls
- Authors
- Name
- DeepHorus
A new open-source project called Conduct has launched to provide guardrails for large language model (LLM) and Model Context Protocol (MCP) tool calls. It aims to offer a unified policy engine that decides what AI agents can do before they take action. The tool integrates with popular developer environments like Claude Code, Cursor, Copilot, and Codex sessions.
The core of Conduct is its policy-first approach, acting as a control plane for AI agents. A single policy can govern blocking, warning, auditing, or injecting actions for every LLM call, shell tool, and MCP invocation. This applies uniformly whether the agent is scheduled, run by a developer, or part of a chat session on the platform.
Conduct operates across three key enforcement surfaces: a CLI hook for developer tools, the MCP layer for tool invocations, and a router for any SDK-speaking LLM call. Each request passes through a Guard check for policy, budget, and audit before reaching upstream providers like Anthropic or OpenAI. The system includes features for audit trails, such as signed configurations and hash-chained logs.
The project offers over 20 compliance packs out of the box, including OWASP, SOC 2, HIPAA, PCI DSS, and the EU AI Act. There are also 22 pre-built playbooks for tasks like issue-to-PR workflows, code review, and incident response. A free Discovery mode provides read-only visibility into AI actions for 14 days without requiring any policy configuration.
For deployment, users can self-host using Docker Compose for a local setup or use Kubernetes templates. There is also a hosted version at conductai.ai, with a free tier for Discovery and paid tiers for enforcement features. The project includes comprehensive documentation, such as a security policy and a detailed threat model.
This release addresses a growing need for oversight as AI agents gain more autonomy in software development. By providing auditable controls, Conduct seeks to help organizations manage risk and comply with emerging regulations. The open-source availability is expected to encourage adoption and further development of agent governance frameworks.
