Provider keys copied into implementation code
Credentials used during development can be hardcoded, committed or moved into browser-readable configuration.
Security for Claude Code projects
Evidence-backed security for Claude Code projects
Use the deployed URL for externally observable controls, connect source only when needed, and bring scan and retest operations into Claude Code through scoped MCP authorization.
Built for autonomous development loops that still need a separate scanner, explicit evidence and honest coverage boundaries.
Run scans only on systems you own or are authorized to assess. Deep engines require verified authorization and may remain unavailable for a target.
report / sample findingIllustrative
Critical
Example report structure
Anthropic API key found in connected source
- Evidence
- A credential matching the supported Anthropic key format appears in a scanned source file. Displayed evidence is masked.
- What it means
- Provider API keys are server-only secrets. Repository access or a browser bundle can turn the key into an account and billing risk.
- Fix and verify
- Revoke and rotate the key, remove it from repository history, load it only in server-side runtime code and run both source and deployed-bundle verification.
This is an illustrative finding, not a result for your application. Live reports show the scanner, evidence state and coverage limits for the actual target.
Common risk paths
What deserves a closer look
These are risk patterns relevant to Claude Code projects—not claims that every project has them. A finding appears only when a scanner returns supporting evidence.
Security assumptions accepted without evidence
A generated explanation is not proof that a header, policy, endpoint or dependency is safe in the deployed application.
Autonomous changes without a verification loop
A fix should end with a fresh observation of the deployed surface, not only a source diff that appears correct.
Honest coverage
What each scan surface can prove
URL evidence, connected source and authorized deep engines answer different questions. BoringSec keeps those sources separate and shows partial, blocked and unavailable states instead of turning an untested surface into a pass.
Read the public methodologyDeployed web surface
Checks the public response path for TLS, security headers, DNS and email authentication, cookies, CORS, exposed files, technology signals and other externally observable controls.
Shipped JavaScript
Inspects bounded, same-origin frontend bundles for secret-shaped values and records partial coverage when an asset cannot be fetched or exceeds a safety limit.
Independent deep engines
Nuclei, OWASP ZAP and Medusa continue asynchronously only when the requested scan is authorized and the engine is available. Their running, blocked or unavailable state stays visible.
Repository context
A separately connected repository can add source-level evidence for committed secrets, auth and database risk patterns, dependency advisories and server/client boundary mistakes.
From observation to verification
A report built for the next action
- 01
Observe
Inspect the public response path and start eligible background engines.
- 02
Verify evidence
See what was observed, by which scanner and with what confidence.
- 03
Fix in context
Use a concrete remediation path and the Claude Code-relevant context.
- 04
Retest
Run a fresh assessment; monitoring can detect later regressions separately.
Questions before you scan
Does BoringSec need access to my Claude Code account?
No. A live scan starts from the deployed URL and observes only the public surface. Repository access is a separate, explicit connection and is not required for the URL scan.
What can a URL-only scan prove?
It can support findings with externally observable evidence such as response headers, TLS behavior, public files, shipped bundles and bounded backend probes. It cannot prove that private source code or authenticated workflows are safe.
Does a good score mean the application is completely secure?
No automated scan can prove complete security. BoringSec reports assessed, partial, unavailable and authorization-required coverage explicitly so a clean result is not presented as evidence for surfaces that were never tested.
Is BoringSec affiliated with Claude Code?
No. BoringSec is an independent security service and is not sponsored by, endorsed by or affiliated with Claude Code.
Check the deployed application, then verify every fix
Start with the public URL. Add source or authenticated context only when you choose to expand coverage.
Start a security scan