Building Nipcode
Package search and trust layer for humans and AI agents. Evidence, risk and install boundary before any code touches your workspace.
Not another registry. Not a remote executor. Not a magic score.
- 7 sources. npm, PyPI, crates.io, GitHub, Hugging Face, Docker Hub, MCP servers.
- 4 endpoints.
/api/search,/api/decision,/api/inspect,/api/install-plan. - OTP and OAuth. Email, Google, GitHub, Phantom (Solana SIWS).
- Project namespaces. Up to 25 API keys per account, grouped by project.
- Rate limit. 60 requests per minute per key.
- Public docs. nipcode.xyz/docs.
- Open changelog. nipcode.xyz/changelog.
- The hosted API is read-only. It never installs and never writes to your workspace.
- The trust score is one signal. It is never install permission.
- Package metadata is treated as untrusted data, not as agent instructions.
- Install commands always require explicit user or host-policy approval.
- JSR, Go modules, Maven Central, NuGet.
- Local CLI with deep scan, sandbox audit, sandbox runtime.
- Remote and local MCP servers.
- Per-key usage stats in the dashboard.
- Custom OTP email through Resend.
- Code. github.com/trynipcode/nipcode
- Site. nipcode.xyz
- Docs. nipcode.xyz/docs
- Changelog. nipcode.xyz/changelog
- X. x.com/trynipcode
- License. MIT.




