0sec: Exploit-backed MCP server for autonomous security testing
0sec, from 0sec Labs, is an open-source MCP server and cybersecurity harness designed to automate evidence-based vulnerability discovery. The tool coordinates autonomous AI agents to probe codebases, web applications, APIs, and package registries and produces findings only when an independent agent reproduces an exploit. Key capabilities include parallel agent hunts, blind verification, strict scope enforcement, and MCP-native access for AI clients. Security researchers and DevSecOps teams who require reproducible, high-confidence penetration testing are the intended users.
What tasks can you actually use it for?
The tool targets actionable security research rather than checklist scanning, and it supports multi-target scanning across web apps, APIs, source repositories, and package ecosystems. It explicitly lists npm, PyPI, and OCI registries as supported package sources. Practical tasks include probing for CORS, SSRF, XSS, SQL injection, and chaining exploits across components using parallel, agentic exploration to reach complex, multi-step attack paths.
How reliable are the reported vulnerabilities?
Reliability is anchored to the "reproduce before trusting" rule: every potential vulnerability must be validated by a separate blind agent before it is reported. The tool generates evidence-backed reports that include protocol attempts, sanitizer crashes, and working proof-of-concept exploits. This verification model changes a finding from an alert to a demonstrable exploit, raising confidence in results compared with scanners that return heuristic scores without reproduction.
What file inputs and operational limits should teams expect?
Deployment runs as a CLI on macOS (Apple Silicon) and Linux (x64/arm64) and requires a connection to a supported LLM provider. The tool functions as an MCP server and integrates with MCP-compatible clients such as Claude Desktop, which means an MCP-capable client and model access are necessary parts of the workflow. Scans require a user-defined scope file because the tool enforces boundaries before tool calls proceed.
Does it fit into existing DevSecOps or research workflows?
Designed for agent-native workflows, the tool supplies programmatic scanning primitives that fit into automated security pipelines and research playbooks. It is open-source and hosted for inspection, which helps teams audit behavior and adapt scripts. The application is described as well-regarded within the AI security community for its MCP integration and exploit-backed approach, making it more suitable to teams that combine LLM infrastructure with security expertise.
Clear fit for expert teams who need exploit-level assurance
0sec is a rigorous choice for security teams that require reproducible exploit verification and have the operational skills to run a CLI and manage model access. It demands infrastructure and security expertise to configure scope files and to supervise autonomous agents, so organizations without LLM integration or experienced analysts should evaluate whether they can support those operational requirements before adopting it.
