A second opinion on everything your agent ships.

It usually runs, and it usually ships with the same handful of holes: keys committed to source, auth routes with no guard, missing rate limits, imports of packages that do not exist. This is a scan for people shipping apps their agent mostly wrote, and it finds those before your users do.

No writes. The scan reads your repo and your live site and returns findings. Any fixing happens in your editor, by you or your agent.

Yes. Connect the GitHub App and access is scoped to read-only. Source is analysed only for the length of the scan, then dropped.

JavaScript and TypeScript, Python, Go, Ruby, Java, PHP and Rust, with framework-aware checks for React, Next.js, Express, Fastify, Django, Flask, FastAPI, Rails and Spring. Anything the site-side checks can reach over HTTP is covered as well.

Paste one line into the agent: connect to vibedoctor using https://vibedoctor.io/mcp/start. It fetches the instructions, installs the MCP server itself, and reports back in the chat you already have open. No keys to copy, no config file to edit.

Every scan verifies each dependency against the registry, so an import of a package that does not exist is flagged with its file and line. It is one of the most common AI-generated defects: 4% of the repositories scanned in production had at least one.

There is real overlap. VibeDoctor runs the same classes of engine they do: a SAST pass, secret detection, dependency CVE scanning, and each language's own linter and type checker (opengrep, gitleaks, trivy, ruff, biome, pyright, tsc, clippy, rubocop, phpstan). What it adds is your deployed site scanned alongside the repo (Lighthouse, security headers, SSL, SEO, exposed files, console errors), checks aimed at AI-generated code such as imports of packages that do not exist, and a graph of your codebase. It also runs from inside your agent over MCP, so findings arrive in the chat where the code was written.

A four-level visual explorer for your codebase: modules, files, symbols and the dependencies between them. It shows what your agent actually built - which functions call what, where complexity hides, and what breaks if you touch something. It updates with every scan.

Yes. The free plan covers one project, one scan a day, and the full report, with no card required. Paid plans add more projects, a scan on every push, and PR review.