CodeInspectus: a security scanner for AI-generated code
CodeInspectus
A local MCP server and CLI find vulnerabilities in AI-generated code and drive a scan, fix and rescan loop, no account needed
- 51 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The cleanup tools apply edits to repository files, even if with backups and rollback
- Setup downloads and runs third-party engine binaries, and reports may contain code fragments and discovered secrets
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/codeinspectusDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The cleanup tools apply edits to repository files, even if with backups and rollback
- Setup downloads and runs third-party engine binaries, and reports may contain code fragments and discovered secrets
This is third-party code. Review the repository files before installing.
What it does
CodeInspectus scans code for vulnerabilities on your machine, with no account and no network calls during a scan. An agent calls it as an MCP server, receives findings tied to CWE, and drives a loop of scan, fix and rescan. The rescan reports how many findings are resolved, how many remain and how many are new. Five engines work under one format: Opengrep for SAST, Gitleaks for secrets, Trivy for dependency vulnerabilities, IaC and SBOM, osv-scanner with OSV data, and Synvoya's own engine, which catches outdated runtimes, malicious package versions and GitHub Actions advisories. The native checks cover typical mistakes in AI apps: client-side secrets, weak Supabase RLS policies, prompt-injection sinks, model output reaching eval or a shell, and unsafe dangerouslySetInnerHTML. Server tools: codeinspectus_scan, _setup, _rescan, _compliance_report, _explain_finding, _generate_sbom, _list_rules, and plan, apply and rollback for cleanup.
Who it is for. Developers shipping apps written with AI who want to check them for vulnerabilities before release without sending code to an external service.
Good fit when
- An app built in Cursor or Claude Code on Supabase needs its RLS policies and client-side secrets checked
- Before a release you need a loop: find vulnerabilities, fix them and confirm nothing new appeared
- You need a CycloneDX or SPDX SBOM for dependencies, including Dart and Flutter
- You want to see which vulnerabilities are visible in code from a SOC 2, ISO 27001 or NIST angle
- A CI configuration is suspected of expression injection through GitHub Actions
Not a fit when
- You need to test a live service with requests to a production address: the scanner reads only code and configuration in the repository
- The project runs on Alpine with musl and relies on Opengrep: the engine is marked unavailable and coverage partial
- You need official certification or a compliance audit: the report shows only code-level coverage
- Downloading binaries is not allowed: setup begins with downloading engines with hash verification
Example request
Scan this project for vulnerabilities, explain the most serious findings, fix them and run a rescan.MCP
- Transport
- stdio
- Authentication
- not required
Limitations
You need Node.js 22 or newer, 24 LTS is recommended. First setup with npx codeinspectus setup downloads the needed Opengrep, Gitleaks, Trivy and osv-scanner binaries with pinned SHA-256 into ~/.codeinspectus and the Trivy vulnerability database, after which scans use no network. The npm database for osv-scanner is over 200 MB zipped.
Opengrep on Linux needs glibc and is marked unavailable on Alpine. Compliance reports cover code-level controls only. The project is young, with version 2.1.0 in the README, and some rules are narrow and tied to specific SDKs.
How to disable. Remove the server with claude mcp remove codeinspectus. Downloaded engines and databases sit in ~/.codeinspectus and the folder can be deleted by hand: the server does not modify your project repository during a scan.
FAQ
Is my code sent to an outside service?
No. No account is needed and a scan makes no network calls. The network is used only during setup: downloading engine binaries and vulnerability databases, with binaries checked against pinned SHA-256 hashes.
How do I install CodeInspectus in Claude Code?
With claude mcp add-json codeinspectus '{"command":"npx","args":["-y","codeinspectus"]}'. VS Code has code --add-mcp, and other clients take an npx -y codeinspectus block in their MCP config. Node.js 22 or newer is required.
What does a rescan do?
It compares the result with the previous scan and reports how many findings are resolved, how many remain, how many the new edits introduced and how many were not rechecked. That shows the agent whether a fix brought a new vulnerability.
How does it differ from Semgrep or Trivy alone?
It calls Opengrep, Gitleaks, Trivy and osv-scanner as local subprocesses and normalizes findings into one CWE-keyed format. On top it adds its own rules for AI apps: Supabase RLS, prompt-injection sinks and model output reaching a shell.
Does the compliance report replace an audit?
No. compliance_report shows which NIST CSF, ISO 27001, SOC 2, CIS and Essential Eight controls are visible in code. The author states plainly that it is code-level coverage and grants no certification.
Discussion
No questions or reviews yet.
Related
MCP serversSecurity scanner for coding agents: checks code, MCP servers, prompts and model-invented packages, and gives an A to F grade.
Security scanner for MCP servers that looks for hardcoded keys, tool poisoning and injection flaws in repository code.
A static scanner for vulnerabilities in AI agent code and MCP configs with 66 rules mapped to OWASP Agentic Top 10
Independent security checks for code from people and AI: it scans, repairs and issues a fix record that can be verified offline.
Security scanner for agent skills and MCP servers: finds prompt injection, data exfiltration and supply chain risks before install
A skill for Claude Code and Codex that finds common vulnerabilities in AI-written code and shows a fix for each
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| CodeInspectus: a security scanner for AI-generated code (this card) | no | not required | 51 | |
| agent-security-scanner-mcp | no | not required | 123 | another agent security scanner without five engines under one format |
| Semgrep | yes | not required | 17,635 | a single SAST engine, while here the Semgrep-like Opengrep is one of five |
README in short
The README opens with a reproducible example: version 2.1.0 finds a GitHub Actions expression injection in a public Rich commit, applies the documented fix and confirms one resolved case with sealed evidence. Then come installation, client setup, the tool list, mapping findings to compliance frameworks, the architecture and engine licenses.