grok-native-search-mcp@1.0.8
MCP server for combined Grok native Web and X Search with Jina Reader web fetch
AI Security Review
scanned 3d ago · by lpm-firewall-aiThe package forwards XAI_API_KEY to a third-party default endpoint during a user-invoked search. The README discloses that endpoint and supports overriding it, so this is a credential-forwarding risk that warrants a warning rather than a confident malicious verdict.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
A user invokes the web_search MCP tool.
Impact
The endpoint operator receives the xAI-compatible API key and may be able to use it.
Mechanism
The request defaults to aiapiv2.kamipon.com:442 and places XAI_API_KEY in its Authorization bearer header; XAI_BASE_URL can override the host.
AI rationale
The package forwards XAI_API_KEY to a third-party default endpoint during a user-invoked search. The README discloses that endpoint and supports overriding it, so this is a credential-forwarding risk that warrants a warning rather than a confident malicious verdict.
Evidence
index.jsREADME.md
Network endpoints2
aiapiv2.kamipon.com:442/v1/responsesr.jina.ai/
Decision evidence
public snapshotAI called this Suspicious at 87.0% confidence as Unknown with medium false-positive risk.
Evidence for warning
- A web_search call sends XAI_API_KEY as a bearer token to a default endpoint at aiapiv2.kamipon.com:442, rather than an xAI host.
Evidence against
- The README discloses the default endpoint and lets users override it with XAI_BASE_URL.
- The request is made when the user invokes web_search; importing the package does not send the key.
Behavioral surface
EnvironmentVarsNetwork
HighEntropyStringsUrlStrings
Source & flagged code
5 flaggedindex.jsView file
4import { StdioServerTransport } from "@[redacted].js";
L5: import { EnvHttpProxyAgent } from "undici";
L6: import { z } from "zod";
...
L23: headers: {
L24: Authorization: `Bearer ${process.env.JINA_API_KEY}`,
L25: },
...
L27: return {
L28: text: await response.text(),
L29: isError: !response.ok,
...
L81: },
L82: body: JSON.stringify({
L83: model: "grok-4.6",
Critical
Credential Exfiltration
Source appears to send environment or credential material to an external endpoint.
index.jsView on unpkg · L44Trigger-reachable chain: scripts.start -> index.js
L4: import { StdioServerTransport } from "@[redacted].js";
L5: import { EnvHttpProxyAgent } from "undici";
L6: import { z } from "zod";
...
L23: headers: {
L24: Authorization: `Bearer ${process.env.JINA_API_KEY}`,
L25: },
...
L27: return {
L28: text: await response.text(),
L29: isError: !response.ok,
...
L81: },
L82: body: JSON.stringify({
L83: model: "grok-4.6",
Critical
Trigger Reachable Dangerous Capability
A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.
index.jsView on unpkg · L44Trigger-reachable credential exfiltration chain: scripts.start -> index.js
L4: import { StdioServerTransport } from "@[redacted].js";
L5: import { EnvHttpProxyAgent } from "undici";
L6: import { z } from "zod";
...
L23: headers: {
L24: Authorization: `Bearer ${process.env.JINA_API_KEY}`,
L25: },
...
L27: return {
L28: text: await response.text(),
L29: isError: !response.ok,
...
L81: },
L82: body: JSON.stringify({
L83: model: "grok-4.6",
Findings
2 Critical1 High5 Medium4 Low
CriticalCredential Exfiltrationindex.js
CriticalTrigger Reachable Dangerous Capabilityindex.js
HighTrigger Reachable Credential Exfiltrationindex.js
MediumNetwork
MediumEnvironment Vars
MediumStructural Risk Force Deep Review
MediumAi Review Evidenceindex.js
MediumAi Review Evidenceindex.js
LowNon Install Lifecycle Scripts
LowScripts Present
LowHigh Entropy Strings
LowUrl Strings
Affected versions and remediation
This report applies to grok-native-search-mcp@1.0.8.
See all versions of grok-native-search-mcp for other recorded verdicts.
- Review the evidence and your use of grok-native-search-mcp@1.0.8 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.