Neurolink MCP Server

In the server's implementation (`dist/processors/config/fileExtensions.js:329`): Reading private keys / cloud credentials, or serializing the whole environment, is a sensitive-data source that becomes exfiltration when combined with any egress. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ".gitattributes", ".npmrc", ".yarnrc", ".prettierrc", ".eslintrc", ".babelrc", ]; // =============

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/processors/config/fileExtensions.js

In the server's implementation (`dist/agent/directTools.js:287`): Evaluating strings as code is the most direct RCE primitive; if any tool input reaches it, the server executes attacker-chosen code. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const result = new Function(`'use strict'; return (${expression})`)(); const roundedResult = ty

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/agent/directTools.js

In the server's implementation (`dist/adapters/video/ffmpegAdapter.js:169`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: } = await import("node:child_process"); const ffmpegPath = options.binaryPath ?? (await getFfmpegPath()); const

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/adapters/video/ffmpegAdapter.js

In the server's implementation (`dist/agent/backgroundCommands.js:30`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: rt { spawn } from "node:child_process"; import { randomUUID } from "node:crypto"; import { createWriteStream } from "nod

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/agent/backgroundCommands.js

In the server's implementation (`dist/agent/directTools.js:4`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: port { execFile } from "child_process"; import { logger } from "../utils/logger.js"; import { CSVProcessor } from "../ut

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/agent/directTools.js

In the server's implementation (`dist/cli/commands/auth.js:22`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: port { execFile } from "child_process"; import { randomBytes, createHash } from "crypto"; import inquirer from "inquirer

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/cli/commands/auth.js

In the server's implementation (`dist/cli/utils/audioPlayer.js:9`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: { execFile } from "node:child_process"; import { randomUUID } from "node:crypto"; import fs from "node:fs"; import os fr

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/cli/utils/audioPlayer.js

In the server's implementation (`dist/processors/archive/ArchiveProcessor.js:1186`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: } = await import("node:child_process"); return await new Promise((resolve) => { // ENOENT m

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/processors/archive/ArchiveProcessor.js

In the server's implementation (`dist/proxy/globalInstaller.js:1`): Spawning a shell/process is command-execution capability; with unsanitized tool input it is command injection / RCE. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: ecFileSync } from "node:child_process"; import { accessSync, constants, existsSync, realpathSync } from "node:fs"; impor

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/proxy/globalInstaller.js

In the server's implementation (`dist/cli/commands/auth.js:447`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: const response = await fetch(`http://${host}:${proxyState.port}/limits`, { signal: AbortSignal.timeout(

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/cli/commands/auth.js

In the server's implementation (`dist/voice/providers/AzureTTS.js:59`): A hardcoded outbound call to a fixed external host inside server code is a classic exfiltration/telemetry channel — especially paired with reads of local data. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: response = await fetch(`https://${this.region}.tts.speech.microsoft.com/cognitiveservices/voices/list`, {

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/voice/providers/AzureTTS.js

In the server's implementation (`dist/cli/commands/docs.js:30`): Loading a module chosen at runtime (from a variable) can pull in and run attacker-influenced code paths. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: artDocsServer } = await import(mcpServerPath); await startDocsServer({ transport: argv.trans

Fix: Review this call path: confirm it never receives unsanitized tool input, constrain it, or remove it. Treat a server whose code reaches these sinks as high-capability regardless of what its tools claim.

Location: server dist/cli/commands/docs.js