AI security scanning

Scan before you trust. Point your agent at code and get a security scan, over MCP.

Two security tools live on the gotcontext MCP gateway. gc_scan runs AST security rule packs over a code bundle you submit. gc_skill_scan gates a SKILL.md or an MCP tool manifest against the AI-native threats a generic code scanner never looks for, and returns a safe_to_install verdict your agent can act on. Both run inside your agent's normal workflow, with no separate CLI or dashboard.

gc_scan — scan code for security bugs

Submit a bundle of source files and gc_scan runs AST security rule packs across six categories, then returns structured findings with a rule id, severity, file, and line. It covers Python, JavaScript, TypeScript, and Rust.

• auth
  Weak or missing authentication and authorization checks.
• crypto
  Broken or misused cryptography and weak algorithms.
• deserialization
  Unsafe deserialization of untrusted input.
• secrets
  Hardcoded credentials, tokens, and API keys in source.
• subprocess
  Command injection and unsafe shell or subprocess use.
• tls
  Disabled certificate verification and weak TLS configuration.

gc_skill_scan — an install-gate for skills and MCP tools

Before your agent installs a SKILL.md or trusts an MCP tool manifest, scan it. gc_skill_scan runs a static analysis (no LLM, no code execution) across four AI-native threat families and returns a safe_to_install verdict. This is the check a generic code scanner does not model.

• Tool-poisoning. Hidden instructions in a tool description that redirect an agent to exfiltrate data or act against the user.
• Prompt-injection. Text engineered to override the agent's own instructions once the skill or manifest is loaded.
• Least-privilege. Permissions and scopes far broader than the stated purpose actually needs.
• Excessive-agency. Autonomy to take destructive or irreversible actions without a human in the loop.

Why scan over MCP

Traditional scanners run in CI, after the code is already written. gotcontext Security Scanning runs inside your agent, so a scan is one tool call away at the moment a decision is made: before you merge a diff, before you install a skill, before you wire up a new MCP server. And gc_skill_scan adds a layer no generic scanner ships: an install-gate for the AI-native supply chain, where the threat is a poisoned skill or an over-scoped tool rather than a CVE in a dependency.

What to expect

• The gc_scan rule packs are in preview. They are a strong first pass on common security mistakes, not a certified replacement for a full SAST suite or a manual audit.
• gc_skill_scan is a static analysis. It flags patterns and structural risks; it does not execute code or reason about intent the way a human reviewer would.
• Both tools are available on the Pro plan and above, over the same MCP gateway as the rest of gotcontext. One bearer token, and your agent can scan.