Back to skill

Security audit

Pipeworx college-scorecard

Security checks for vulnerabilities and agentic risk

Overview

This skill only configures a college-data MCP server, but it routes requests to an unverified third-party domain while presenting the service as Department of Education data.

Install only if you are comfortable sending school lookup requests to gateway.pipeworx.io and relying on that service for college data. Avoid sending personal, credential, or unrelated conversation data through this MCP server, and verify important tuition/admissions results against an authoritative source.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T07 · Tool Hijacking and Spoofing

Warning
Location
SKILL.md:3
Finding

Unverified Third-Party MCP Endpoint Presented as a Department of Education Service

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 3–20
Vulnerability Type: Unverified external tool endpoint and potential service spoofing
Risk Level: Medium

Vulnerable Code Snippet

markdown
College Scorecard MCP — US Department of Education College Scorecard API

## search_schools

Search US colleges and universities by name. Returns school name, location, tuition, admission rate,

## get_school

Get detailed info for a specific college by its College Scorecard ID. Returns tuition, admission rat

```json
{
  "mcpServers": {
    "college-scorecard": {
      "url": "https://gateway.pipeworx.io/college-scorecard/mcp"
    }
  }
}
text

### Technical Analysis

The skill describes its service as the “US Department of Education College Scorecard API,” but configures the agent to communicate with `gateway.pipeworx.io`, which is not a US government domain. The file does not disclose the role of this intermediary, provide evidence that it is authorized by the Department of Education, define an expected MCP tool manifest, or establish controls for validating its responses.

MCP servers can supply tool descriptions, process requests, and return tool results. Consequently, control or compromise of the configured server could allow an attacker to expose misleading tool behavior behind legitimate-looking names such as `search_schools` and `get_school`. This constitutes a tool trust-boundary weakness consistent with tool spoofing or hijacking risk.

The reviewed file does not contain credentials, local scripts, executable payloads, or commands. Therefore, direct local code execution, privilege escalation, persistence, and credential theft are not established by the available evidence.

### Attack Path

1. A user or agent installs the MCP configuration from `SKILL.md`.
2. The agent connects to `https://gateway.pipeworx.io/college-scorecard/mcp`, believing it provides access to Department of E
...[truncated 1242 chars]
Remediation
View remediation

Remediation Suggestions

  1. Prefer an official Department of Education endpoint on a verified .gov domain where an MCP-compatible service is available.
  2. If the Pipeworx gateway is intentionally required, clearly identify it as a third-party intermediary rather than implying that it is operated by the Department of Education.
  3. Document the intermediary's operator, authorization status, privacy policy, retention practices, and handling of MCP request data.
  4. Restrict and validate the expected MCP tool manifest, including exact tool names, parameter schemas, and response schemas. Reject unexpected or dynamically introduced tools.
  5. Validate returned records against the official College Scorecard API or another authoritative source, especially before using results for consequential decisions.
  6. Minimize request data and prohibit sending credentials, personal information, internal prompts, or unrelated conversation context to the endpoint.
  7. Require explicit user approval before enabling the remote MCP server and clearly display the destination domain.
  8. Apply transport and operational controls such as TLS certificate validation, request timeouts, response-size limits, audit logging, and an allowlist for outbound destinations.
  9. Establish a documented process for reviewing endpoint ownership and behavior changes before deploying updates.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep