CVE Posture76/100
Some dependency vulnerabilities were found. Patch or upgrade packages to raise this score.
Analyzing package security…
Analyzing package security…
Package trust score
This server scores 50/100 (grade C). The biggest gaps are attack history and response hygiene — fixing those will raise your score fastest. Evidence is partial (0% coverage): Behavioral deep-scan produced no runtime data (install/egress hook unavailable). Lower coverage caps the score — re-run with runtime layers enabled for a stronger grade. Top recommendation: improve attack history score (+~5 pts).
This server scores 50/100 (grade C). The biggest gaps are attack history and response hygiene — fixing those will raise your score fastest. Evidence is partial (0% coverage): Behavioral deep-scan produced no runtime data (install/egress hook unavailable). Lower coverage caps the score — re-run with runtime layers enabled for a stronger grade. Top recommendation: improve attack history score (+~5 pts).
| Category | Score | Weight | Points |
|---|---|---|---|
| CVE Posture | 76/100 | 15% | +11 |
| Supply Chain & Malware | 70/100 | 20% | +14 |
| Supply Chain | 76/100 | 10% | +8 |
| Authentication | 55/100 | 8% | +4 |
| Transport Security | 60/100 | 5% | +3 |
| Tool Capability | 92/100 | 15% | +14 |
| Attack History | 45/100 | 15% | +7 |
| Response Hygiene | 50/100 | 10% | +5 |
| Mastyf AI Protection | 50/100 | 2% | +1 |
Some dependency vulnerabilities were found. Patch or upgrade packages to raise this score.
Pick a layout and copy markdown, HTML, RST, BBCode, or AsciiDoc for your README.
Fix the issues above, then run a deep scan or publish from your mastyf.ai proxy for a maintainer-verified badge.
Trust score: 50/100 (live scan)
No known CVEs
Verified publisher
Malware/supply-chain heuristics raised flags — review the signals before trusting this package.
Supply-chain signals need review — verify package name and publisher before trusting.
Only basic API-key auth is configured. OAuth or mTLS is stronger for production.
Uses local stdio transport (low network exposure) but no wire encryption if exposed remotely.
Tool surface is mostly read-only or low risk.
No attack data observed — static review only or no probe coverage.
Tool responses are not scanned for leaked secrets or PII — enable Response DLP.
Not proxied through Mastyf AI — no runtime policy enforcement or audit trail.
Plain-language findings from the security scan — fix these to improve your score.
The scan produced evidence from 0% of its runtime layers. Behavioral deep-scan produced no runtime data (install/egress hook unavailable) — so the differentiated runtime checks could not be verified and the score is capped until they run.
How to fix: Re-run the scan with a reachable live server and a successful behavioral install so egress and attack-probe layers produce data.
@modelcontextprotocol/sdk: This affects @modelcontextprotocol/sdk (direct dependency) and is reported by OSV/NVD.
How to fix: Update the affected package to the latest patched release.
@modelcontextprotocol/sdk: This affects @modelcontextprotocol/sdk (direct dependency) and is reported by OSV/NVD.
How to fix: Update the affected package to the latest patched release.
@modelcontextprotocol/sdk: This affects @modelcontextprotocol/sdk (direct dependency) and is reported by OSV/NVD.
How to fix: Update the affected package to the latest patched release.
zod: This affects zod (direct dependency) and is reported by OSV/NVD.
How to fix: Update the affected package to the latest patched release.
The server accepts tool calls without verifying who is calling. In shared or remote setups, attackers could invoke file or system tools directly.
How to fix: Add OAuth 2.1, API keys, or mTLS before exposing this server beyond localhost.
The server talks over local stdin/stdout — low network risk when run on the same machine, but no encryption if tunneled remotely.
How to fix: Keep stdio for local dev; use HTTPS/mTLS or Mastyf AI proxy for remote agents.
Tools like "delete_agent", "delete_annotation_issue" can execute commands, delete files, or change system state. A compromised agent could abuse them.
How to fix: Block or shadow these tools in YAML policy until explicitly approved per workflow.
Tools such as "create_agent", "create_annotation_issue", "update_annotation_issue" can create or modify data. Review whether agents need write access.
How to fix: Use read-only tool subsets where possible; enable cost and audit limits in policy.
The runtime probe tried to start the server and probe it with attack payloads, but could not establish a live MCP connection: Process exited with code 1. No attack observations were collected.
How to fix: Confirm the package starts an MCP server (some expose it behind a subcommand) and re-run a live probe.
Improve attack history score