Ship Safe

What it does

Ship Safe is mainly a command-line security scanner for code written with AI help. Its `mcp` subcommand starts a stdio MCP server, so a coding assistant can run the same checks on a project while you work on it, without leaving the chat.

The server has seven tools in version 11.0.0, although the README does not list them: the list is in cli/commands/mcp.js. `scan_repo` runs the full set of security agents (injection, auth, secrets, supply chain, LLM and MCP configuration and more) over a directory and returns findings with severity, a score and fix advice. `scan_secrets` and `analyze_file` run the pattern and entropy checks for leaked keys and credentials. `get_findings` reads a JSON report saved earlier, `get_checklist` returns a launch-day security checklist, `suppress_finding` marks a false positive with an ignore comment in the source file, and `ship_safe_status` confirms the connection.

The core checks are deterministic and run on your machine with no account or API key. `scan_repo` caps the findings it returns at 200 by default (`maxFindings` changes it; `outputFile` keeps the full report), and runs agents six at a time with a 30-second timeout each. Its optional `llm` flag adds a model-backed review of critical and high findings through whichever provider key it finds in the environment, with a 50-cent budget per run set in the code.

What it needs

Node.js 22.13 or later on the 22.x line, or Node.js 24 and later (Node 18 and 20 are not supported), and npx. No account or API key for scanning. The optional `llm` deep analysis in `scan_repo` needs a model provider key such as ANTHROPIC_API_KEY in the environment.

Limits

Only `scan_repo`, `get_findings`, `scan_secrets`, `analyze_file`, `get_checklist`, `suppress_finding` and `ship_safe_status` are exposed over MCP; the CLI's investigate, red-team, agent fix, trust and capabilities commands are not tools. The server does not intercept the assistant's actions; it checks only when a tool is called, as its status tool says. `scan_repo` returns at most 200 findings unless `maxFindings` is raised. Writes (`suppress_finding`, `outputFile`) are refused for paths outside the directory the server was started in. Its makers say it complements, not replaces, CodeQL, Gitleaks and Trivy, and does no interprocedural taint analysis like CodeQL.

Who it suits

Suits developers who use a coding assistant and want a quick, local security pass over a repository, especially over AI agent, prompt and MCP configuration that general scanners skip. It needs only Node, and it works offline unless the `llm` option is used. It is a third-party tool, not tied to any vendor's product. Teams needing deep data-flow analysis or a CVE database should keep a dedicated scanner alongside it, as its own README advises. The separate server shipsafe-independent-security-verification (from ship-safe.co) is a different product with a similar name.