MCP Registry

code-engine-mcp-server

4606af71-d168-5f24-aab4-3e4bd3d38f86 | Store v1.7.8
61/ 100
MEDIUM risk
No change since v1.5.0
AI review verdict
Review before use

The AI review rates the findings as likely false positive, but the risk score (61/100) still counts them.

Analysis record

Analysed
1 month ago 1 October 2026
Version
v1.7.8
Artifact
SHA256 41b228dd…b6081e
Detail
Full list

Is code-engine-mcp-server safe?

This is an MCP server that lets an AI assistant deploy containers to IBM Cloud Code Engine from tools like Cursor and Claude. It declares no special permissions and no network endpoints of its own. The only access it needs is the IBM Cloud API key you give it so it can reach IBM's API on your behalf.

Every finding on the scan points at the same place: build/index.js, where the axios HTTP client is bundled into the compiled output. The titles look like NET-AXIOS-build/index.js-2602 and NET-AXIOS-build/index.js-2818. Axios is a common library for making web requests, and a cloud deployment tool needs one. If those hits were real, they would mean the package is contacting somewhere it should not, but they are the library's own code sitting inside a single compiled file, not a hidden destination.

The scan found no tool-poisoning patterns, which is the main thing to check on an MCP server: nothing in the tool descriptions tries to instruct the AI behind your back. It also found no code reading your SSH keys, AWS credentials, or Kubernetes config. The one secret the server does need, an IBM Cloud API key, is read from the environment, which is how every cloud tool works.

Evidence ledger

Ranked by severity · findings with a source location link to the code viewer

150 findings
Showing 25 of 150 · highest severity first

Finding Categories

148
Network

MCP Server Analysis

MCP servers expose tools and resources to AI assistants. Unlike browser extensions, they run as standalone processes with direct system access. Tool definitions are analyzed for prompt injection, data exfiltration, and tool poisoning patterns.

AI Security Report

AI Security Review

Evidence context: threat category none; evidence quality moderate.

The 148 network detections in this package all sit inside one compiled file and all carry the same library prefix: NET-AXIOS-build/index.js-2602, NET-AXIOS-build/index.js-2818, NET-AXIOS-build/index.js-2679, and the rest of the run between lines 2679 and 2818 of build/index.js. Axios is the JavaScript HTTP client. That tight cluster of line numbers is a bundled copy of the library, not a sequence of hand-written calls to servers the package never mentions. Nothing else fired. There are no malware signatures, no credential-access matches, no obfuscation hits, and no secret findings.

Tool poisoning is the threat that matters for an MCP server, and there is none. The scanner returned zero tool-poisoning findings, which means no tool description in this package carries instructions aimed at the model. No "do not tell the user", no invisible Unicode directives, no XML instruction tags, no attempt to have the agent read data and route it somewhere else. The server registers tools that let an assistant deploy containers to IBM Cloud Code Engine. Those descriptions are written for the developer, not for the model to obey.

Credential handling is narrow and expected. The package produced no secret findings and no credential-access findings, so nothing reaches for .ssh, .aws/credentials, or .kube/config. A Code Engine deployment needs an IBM Cloud API key, which this server reads from the environment the same way every cloud MCP server does. Reading process.env for a documented API key is configuration. It turns into exfiltration only when it pairs with a network call to an undocumented destination, and the endpoints list here is empty.

The strongest counterargument is scope. A server that can deploy containers can run arbitrary workloads under the user's IBM Cloud identity, and that is a lot of authority to hand an agent. It is also the entire advertised purpose, stated in plain text on the package. Powerful and misbehaving are separate questions, and the findings answer the second one: no hidden directives, no credential harvesting beyond the one key the job requires, no unexpected destination.

The only thing the scanner reacted to was axios shipping inside a compiled bundle, which is ordinary for any package that talks to a cloud REST API. Two hits are marked high severity, but their titles and paths match the same axios pattern as the medium ones, so that severity reflects the rule, not the code.

Key Reasons

  • All 148 network findings are NET-AXIOS detections confined to build/index.js, matching the line range of a bundled axios HTTP client
  • Zero tool-poisoning findings; no hidden AI directives appear in any tool description
  • No credential-access or secret findings targeting .ssh, .aws/credentials, or .kube/config
  • No network endpoints or host permissions declared, so nothing indicates exfiltration to an unknown domain
  • Remaining findings are all one library, so the total reflects bundling volume rather than misbehavior

False Positive Considerations

  • Bundled axios HTTP client inside compiled build/index.js
  • Repetitive NET-AXIOS detections with identical titles and empty descriptions
  • Zero tool-poisoning, secret, ioc, obfuscation, or code-smell findings

Reviewed 2026-10-02 · version 1.7.8; recommended action: suppress false positive; model confidence 85%.

MCP version history

Risk trend by version

7 analyzed versions. Each point is the latest successful scan for that version; failed zero-score scans are hidden. Dates: First seen by Risky Plugins.

Selected
61
Change since first
+14
Change from previous
No change
Versions:
First analyzed version
1.0.6
First seen May 9, 2026
Risk range
47 to 61
Across analyzed versions
Latest analyzed version
1.7.8
First seen Oct 1, 2026
Selected version
medium
Version
v1.7.8
First seen 1 month ago
Risk score
61
Findings
150
Change vs previous
No change

Pick any point on the chart, then open the Source tab to read that version's code.

Source Code Viewer

Source Code Not Available

Source code is not available for this version of the extension.

About This Extension

IBM Code Engine MCP Server - deploy containers to IBM Cloud Code Engine from Cursor, Copilot, Claude, and Cline using natural language

Frequently Asked Questions