CVE-2025-54073 · high
MCP package-documentation server command injection
mcp-package-docs 0.1.27 and earlier.
Update to 0.1.28 or later and use argument-safe process APIs rather than shell interpolation.
CapitalGuard Observatory · Tool record
MCP-Connected AI Assistants
Multiple providers · connected assistant
MCP-connected assistants can discover resources and call tools exposed by local or remote servers, creating a reusable bridge between AI and files, APIs, databases, commands, and business systems.
Public-Control Score
N/A
Implementation-dependent protocol
MCP is a protocol and ecosystem, not one deployed product. A single score would hide differences between clients, servers, transport, authorization, and operator policy.
Scored dimensions
The documented control surface.
This record is reviewed against the same five-dimension rubric used across every scored product.
Read MethodologyA protocol needs implementation-level evidence.
Score the client, server, authorization method, transport, tool definitions, and operator policy together. A protocol-wide number would be false precision.
Documented access points
1
MCP resources and prompts
2
local stdio server processes
3
remote tools, OAuth scopes, APIs, and downstream services
Settings to verify now
1
Server origin, command, and transport
2
OAuth scopes, token audience, and consent
3
Filesystem, network, session, logging, and downstream permissions
Linked security signals
Version checks that should not wait.
CVE-2026-40576 · critical
Remote MCP spreadsheet server path traversal
excel-mcp-server 0.1.7 and earlier.
Update to 0.1.8 or later, bind remote services narrowly, authenticate requests, and validate resolved paths against the approved root.
CGO-RESEARCH-2026-05 · informational
AgentSecBench formalizes three enforceable agent-security boundaries
Agentic systems that combine untrusted content, private context, and action-capable tools.
Project trusted provenance, restrict capabilities before generation, and validate outputs before execution.
Linked evidence
Ten practical MCP-Connected AI Assistants checks.
Private file access
MCP-Connected AI Assistants Private File Access: What to Check
Credential exposure
MCP-Connected AI Assistants API Key Exposure: Check Before You Share
Client confidentiality
MCP-Connected AI Assistants Client Data Safety for Freelancers
Prompt injection
MCP-Connected AI Assistants Prompt Injection: A Practical Defense
Connector permissions
MCP-Connected AI Assistants Connector Permissions: Review the Real Scope
Command execution
MCP-Connected AI Assistants Command Execution: Keep Control
Unsafe generated code
MCP-Connected AI Assistants Generated Code Risks: Review Before You Run
History and sharing
MCP-Connected AI Assistants Chat History and Shared Links: Privacy Check
Accidental oversharing
MCP-Connected AI Assistants Sensitive Data: What Not to Paste
Autonomous actions
MCP-Connected AI Assistants Autonomous Actions: Approval Safety Guide