MCTS: local-first security scanner for MCP server integrations
MCTS (Model Context Threat Scanner) from MCP Audit is a local-first security auditor for Model Context Protocol servers, built to find threats in AI-agent integrations before deployment. It inspects server tool surfaces, flags permission and logic abuse, and generates machine-readable audit output for engineering review and pipeline gating. The tool targets MCP server authors, platform engineering teams, and security researchers who need protocol-focused pre-deployment assurance.
What tasks can you actually use it for?
MCTS maps the threat surface that AI agents expose via MCP servers and surfaces complex misuse scenarios. It performs both static discovery and live inspection of available tools, and it looks specifically for multi-step exploits where otherwise harmless tools are combined into a malicious sequence. Detection covers permission abuse, secret exposure, tool poisoning, and supply-chain vectors, presented as findings teams can triage.
How effective is it at finding chained threats and logic flaws?
The tool implements chained-risk analysis that identifies risks emerging only when tools interact, a capability the developer highlights as distinct from generic scanners. Outputs include auditable risk scores for individual tools and server configurations, giving teams a traceable basis for triage. Users note it catches complex logic flaws that conventional scanners often miss, making its findings particularly relevant for agentic AI integrations.
What inputs, outputs, and environments does it work with?
MCTS is a Python-based CLI designed to run locally on Windows, macOS, and Linux in any environment that supports MCP servers. Report export formats include JSON, SARIF, and HTML, enabling automated consumption by other tools. The scanner supports both offline analysis and live server inspection, so it can operate inside isolated networks without sending data to external services.
Is it practical to integrate into developer workflows and CI pipelines?
The scanner is built for integration with automated pipelines; it supports CI/CD gating and SARIF export for security tooling that ingests standardized findings. Its protocol-native design focuses on MCP-specific threat models rather than repurposed web-API tests, making it a closer fit for platform engineering teams who must validate AI-to-tool integrations prior to deployment.
Best suited to security-aware engineering teams, not non-technical stakeholders
MCTS is a pragmatic choice for MCP server authors and platform teams who manage AI integrations and require protocol-focused audit output. Because it surfaces multi-step logic interactions, teams should assign security-literate reviewers to interpret flagged items and prioritize fixes. Organizations without embedded security expertise should plan for analyst time to validate findings before treating them as deployment decisions.
