@roarkanalytics/sdk-mcp@3.23.0
The official MCP Server for the Roark API
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThis is a real dangerous MCP capability, but the inspected source contains no install-time execution, persistence, stealth, destructive behavior, or autonomous exfiltration. The local execution path is explicit server functionality with a restricted network allowlist, so the evidence does not establish malware.
Static reason
High-risk behavior combination matched malicious policy.; source matched previously finalized malicious package; routed for review; source fingerprint signature matched known malicious package; routed for review; previous stored version diff introduced dangerous source
Trigger
Start the MCP server and invoke its execute tool with code.
Impact
A malicious prompt or untrusted MCP caller could use the enabled execution tool to access environment-held credentials or invoke permitted Roark API operations.
Mechanism
Caller-controlled code execution with broad environment access.
AI rationale
This is a real dangerous MCP capability, but the inspected source contains no install-time execution, persistence, stealth, destructive behavior, or autonomous exfiltration. The local execution path is explicit server functionality with a restricted network allowlist, so the evidence does not establish malware.
Evidence
src/options.tssrc/code-tool.tssrc/code-tool-worker.ts
Network endpoints2
api.stainless.comapi.roark.ai
Decision evidence
public snapshotAI called this Suspicious at 90.0% confidence as Dangerous Capability with low false-positive risk.
Evidence for warning
- The default CLI enables a local code-execution mode.
- The code tool starts Deno with broad environment access and package-directory read access.
- The worker dynamically imports and invokes code supplied through the MCP execute tool.
- An alternate handler sends the Roark bearer token, submitted code, and client environment data to api.stainless.com.
Evidence against
- package.json has no preinstall, install, or postinstall hook.
- The executable runs only when invoked as the MCP server; importing it does not start the server.
- The local sandbox restricts network access to the configured Roark API hostname.
- No package source writes agent configuration, persists itself, or performs destructive actions.
Behavioral surface
ChildProcessDynamicRequireEnvironmentVarsFilesystemNetworkShell
HighEntropyStringsUrlStrings
Source & flagged code
13 flagged · loading sourceinstructions.jsView file
•matchType = normalized_sha256
matchedPackage = @roarkanalytics/sdk-mcp@3.24.0
matchedPath = instructions.js
matchedIdentity = npm:QHJvYXJrYW5hbHl0aWNzL3Nkay1tY3A:3.24.0
similarity = 1.000
summary = normalized source hash matched finalized malicious source
High
Known Malware Source Similarity
Source file is highly similar to a previously finalized malicious package; route for source-aware review.
instructions.jsView on unpkg7exports.getInstructions = getInstructions;
L8: const promises_1 = __importDefault(require("fs/promises"));
L9: const logger_1 = require("./logger.js");
Medium
Dynamic Require
Package source references dynamic require/import behavior.
instructions.jsView on unpkg · L7code-tool.jsView file
52Source sends the broad process environment to a literal external destination.
L52: interfaceType: 'PHONE',
L53: recordingUrl: 'https://example.com',
L54: startedAt: 'startedAt',
...
L155: },
L156: body: JSON.stringify({
L157: project_name: 'roark-analytics',
...
L166: }
L167: throw new Error(`${res.status}: ${res.statusText} error when trying to contact Code Tool server. Details: ${await res.text()}`);
L168: }
...
L214: // Check if deno is in PATH
L215: const { execSync, spawn } = await Promise.resolve().then(() => __importStar(require('node:child_process')));
L216: try {
Findings
2 Critical9 High5 Medium4 Low
CriticalHardcoded Runtime Data Exfiltrationcode-tool.js
CriticalManifest Confusionpackage.json
HighSemantic Analysis Limitedcode-tool.js#virtual:normalized:round1
HighKnown Malware Source Similaritycode-tool.js
HighKnown Malware Source Similaritycode-tool.mjs
HighKnown Malware Source Similaritysrc/code-tool.ts
HighKnown Malware Source Similaritycode-tool-worker.js
HighKnown Malware Source Similarityinstructions.js
HighKnown Malware Source Similaritylocal-docs-search.js
HighKnown Malware Source Fingerprint Signaturecode-tool.js
HighPrevious Version Dangerous Deltacode-tool-worker.js
Affected versions and remediation
This report applies to @roarkanalytics/sdk-mcp@3.23.0.
See all versions of @roarkanalytics/sdk-mcp for other recorded verdicts.
- Review the evidence and your use of @roarkanalytics/sdk-mcp@3.23.0 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.