MCP Risk

Scan report

ai.agentutility/mcp-browser-workflow

This page reports an automated security scan of ai.agentutility/mcp-browser-workflow (npm), scanned 2026-07-28. It rates trust (provenance, transport, known vulnerabilities, instruction integrity) as grade C and capability risk (what the tools can do) separately at low. It does not certify the server is safe to connect; it reports what could be checked from outside.

npm, scanned 2026-07-28, tool surface via package-source

C 68/100 Capability risk: Low permalink

The grade rates trust: provenance, transport, known vulnerabilities and instruction integrity. Capability risk rates what the tools can do and is reported, not penalised – a server built to run shell commands is doing its job (Policy §2.4).

Automated checks

StatusCheckPolicyResult
PASS Listed on the official MCP registry §1.1
Listed on registry.modelcontextprotocol.io as ai.agentutility/mcp-browser-workflow. 1 finding
  • Registry entry: ai.agentutility/mcp-browser-workflow ↗
FAIL Source repository is public and maintained §1.2

Claimed source repository does not exist (or is private).
WARN Published package is consistent and attested §1.3
Minor integrity gaps: package is only 22 days old; no npm provenance attestation. 6 findings
  • Package: @agentutility/mcp-browser-workflow@0.1.3 ↗
  • Ecosystem: npm
  • Age: 22 days
  • Weekly downloads: 219
  • Repo field matches source repo: true
  • Build-provenance attestation: absent
PASS Tool surface is inspectable and proportionate §2.1–§2.4
Package source scanned; no high-risk capability signals. 1 finding
  • Detected capabilities: none
INFO Transport & authentication §3

Local (stdio) server: runs with host-process privileges by design. Transport checks apply to remote servers; scrutiny shifts to §1/§2.
PASS No known vulnerabilities (OSV.dev) §4.1

No advisories on record for @agentutility/mcp-browser-workflow @0.1.3 (npm).
UNVERIFIABLE Server-supplied instructions free of poisoning indicators §5.1–§5.4

No server-supplied instructions or tool descriptions were obtainable to scan.

Verify manually - the scoring model can’t judge these for you

  • §2.3 - Is the tool surface no broader than the server’s stated purpose requires?
  • §3.4 - For stdio servers: which environment credentials does it read, and are they scoped rather than org-wide tokens?
  • §6.1 - Cross this server’s capabilities with your already-connected servers: any read-capability × send-capability pairs need explicit sign-off.
  • §6.2 - Run the connecting agent with sandboxing and egress controls sized to the combined tool surface, not just this server’s.

Methodology: MCP Server Security Policy. Checks and scoring weights are open source. Unfamiliar term? See the glossary.