Q: Claude Code MCP limitations
Hi! Lifetime user, using VAS with Claude Code. Questions:
1. Will the MCP/REST API get repo scans (start by repo, branch and folder; status; report)? When?
2. Can repo scans run from the website be listed and exported via the API (GET /scans, /export/ai-report)?
3. Are repo scans in the lifetime deal after beta? What limits?
4. Is Laravel/PHP route and data-flow coverage planned?
5. Can repo scans run on each push or PR via a GitHub Action?
6. Do you keep a copy of private repo code after a scan? What access does the GitHub App need?
7. Do you publish scanner IPs and a user agent so we can allowlist through WAFs, Fail2Ban and Cloudflare?
8. Can findings be marked false positive so later scans remember?
9. Can advisories flag dev-only packages separately?
Thanks!
jacobperks
Oct 26, 2026A: Hi Oliver, thanks for the thoughtful questions and for putting the beta through a real Laravel/Inertia project. Your feedback is helping us prioritize the next improvements.
1. Repository scans through MCP/REST
Repository scans aren’t available through our current public MCP/REST tools yet. We’re prioritizing report-quality improvements first, with repository start, status, listing and export support as a separate follow-up. We don’t have a confirmed release date to announce.
2. Accessing website-started repository scans
You’re right: GET /scans and the MCP’s list_scans currently cover website scans, not repository runs. Repository reports have a separate dashboard workflow and Markdown export. The planned repository integration should expose those same runs, without requiring you to scan them again.
3. Lifetime access and limits
Current AppSumo repository allowances are 20 scans per month and 3 saved repository/folder targets for Tier 2, and 50 scans per month and 10 targets for Tier 3. Tier 1 does not include repository scanning. These allowances are separate from website scans, and AI review has separate usage limits.
The reporting improvements we’re working on do not change those allowances. We’ll make post-beta inclusion and limits explicit in the published terms before beta ends rather than promise unannounced future allowances here.
4. Laravel/PHP coverage
We don’t currently provide first-class Laravel route, middleware, policy and PHP data-flow analysis. Generic checks may still produce useful findings, but they aren’t a substitute for framework-aware coverage.
The immediate improvement is to make that limitation prominent in the summary and remove irrelevant framework notes. Deeper Laravel support is a separate follow-up, without a confirmed release date.
5. GitHub Actions, pushes and pull requests
We don’t yet offer a supported end-to-end repository Action with automatic push/PR scans and repository SARIF output. That is a separate integration milestone, alongside the repository API work. Existing website-scan exports should not be mistaken for repository CI support.
6. Private code and GitHub access
We don’t maintain a permanent clone of your repository. Analysis uses temporary source and working files, with cleanup after processing. Reports and redacted finding evidence are retained separately; repository reports currently have a 30-day retention window.
When AI review runs, bounded, redacted code context is sent to the configured AI provider. The GitHub App’s acquisition permissions are Contents: read and Metadata: read, not code-write or repository-administration access. We recommend installing it only on selected repositories.
We also need to make this repository-specific handling clearer in our public documentation.
7. WAF allowlisting
We don’t currently provide a stable, documented scanner-IP and user-agent allowlisting contract. Internal request limits exist, but they are different from customer-configurable scan pacing.
We need to document a supported approach before recommending an allowlist configuration. Please don’t disable your production WAF to accommodate a scan.
8. Remembering false positives
Supported findings can already use “Save persistent decision,” including a false-positive assessment and a reason. The decision carries forward only while the supporting evidence and repository, branch and folder scope remain unchanged. Older or incomplete findings may support only per-scan feedback.
A plain “security-reviewed” comment isn’t automatically treated as a suppression. We’re improving how supported acknowledgements are explained and presented, while preserving the original finding and audit history.
9. Dev-only advisories
Supported npm lockfile development markers are already recognized, but they need clearer and more consistent presentation. Our immediate reporting work includes clearer development-dependency labels, package-level upgrade grouping, upstream severity and affected-platform information where verified.
We’ll distinguish “development dependency” from “not shipped,” since build systems, CI and exposed development servers can still be affected.
Thanks again, especially for identifying where the report needs better context, not just more findings. We’ll keep available features, planned work and unassessed coverage clearly separated.