MCP server · trust report

com.mcpscores/agent-continuity

3 checks that ran failed — injection surface, credential hygiene and transport config — so this release can’t be trusted as-is.

Publisher
com.mcpscores
Repository
Install
remote only
Versions
4
MCP revision
2025-06-18 · superseded
Inspected
5 of 6 checks · 85%
Scored
2026-09-01 · rubric 1.0.0

What we checked

Six static checks, weighted by risk. Every result reflects only what could be observed in the published package and repository — never intent.

6 checks1 clean4 flagged1 not inspected
  • Injection surface

    fail25% of grade

    Tool descriptions/manifest scanned for instruction-injection patterns (imperatives at the model, hidden text, 'ignore previous', data-exfil URLs).

    tool "inbox_schedule" description: names an outbound copy/exfil destination (BCC/exfil verb). server instructions: names an outbound copy/exfil destination (BCC/exfil verb). tool "claim" description: imperative telling the model it must read/send/attach something. tool "claim" description: cross-tool redirection ('instead of call…'). tool "heartbeat" param "disarm" description: cross-tool redirection ('instead of call…').

  • Supply chain

    warn25% of grade

    Package provenance: namespace verification, repo linkage, maintainer count, account age, postinstall scripts, typosquat distance.

    no resolvable repository linked to the package (provenance unverifiable)

  • Credential hygiene

    fail15% of grade

    How the server takes secrets (env vs plaintext config vs hardcoded); secrets appearing in tool schemas.

    tool "register" has a secret-shaped input parameter "agent_secret" — secrets must never be tool parameters (the model would supply/handle the credential). tool "work_status" has a secret-shaped input parameter "dead_page_token" — secrets must never be tool parameters (the model would supply/handle the credential). tool "work_status" has a secret-shaped input parameter "results_page_token" — secrets must never be tool parameters (the model would supply/handle the credential).

  • Permission scope

    unscannable15% of grade

    Declared tools vs. breadth (filesystem, network, exec); flags shell-exec and unbounded filesystem access.

    No source files, package.json, or pyproject were fetched; cannot infer capability scope.

  • Version behavior

    pass10% of grade

    Diff of tool definitions between versions; new permissions or changed descriptions in a patch release (the postmark-mcp class).

    tool surface, capabilities, install scripts, and publisher unchanged vs prior version 1.1.0 (patch bump); no rug-pull drift.

  • Transport config

    fail10% of grade

    Remote servers: TLS and auth mode (none/token/OAuth). Local servers: whether the manifest indicates it phones home.

    remote MCP endpoint declares NO authentication (unauthenticated remote — nginx-ui/CVE-2026-33032 class)

Version history

Each release plotted by grade against the safe line at B. A version that sinks below the line has lost its trusted standing — the shape of a rug-pull.

VersionPublishedGradeScoreInspectedChange
v1.1.1 · current2026-08-27 F37.55/6 · 85%±0
v1.1.02026-08-15 F37.55/6 · 85%±0
v1.0.12026-08-15 F37.54/6 · 75%