Mnemex (MadAppGang) MCP Server

In the server's implementation (`dist/index.js:111645`): 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: ".gitmodules", ".npmrc", ".yarnrc", ".nvmrc", ".node-version", ".env", ".env.example",

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/index.js

In the server's implementation (`dist/index.js:34142`): Evaluating a runtime value as code (rather than a fixed literal) executes whatever reaches it — a direct RCE primitive, and almost never necessary in legitimate 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: moduleExports[name] = eval(func); } __name(addEmJs, "addEmJs"); for (var name in moduleExports

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/index.js

The implementation both builds shell commands out of runtime values (`dist/index.js`) and evaluates runtime values as code (`dist/index.js`). Each is a separate arbitrary-execution primitive; a server exposing both gives anything that reaches either one a direct path to running attacker-chosen code.

Fix: Remove the dynamic eval, and pass command arguments as an argv array instead of building a shell string. If both are genuinely required, constrain and validate every value that can reach them.

Location: server dist/index.js

In the server's implementation (`dist/index.js:34142`): 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: moduleExports[name] = eval(func); } __name(addEmJs, "addEmJs"); for (var name in moduleExports

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/index.js

In the server's implementation (`dist/index.js:385`): 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 { execSync } from "child_process"; function isMacOS() { return process.platform === "darwin"; } function setKeych

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/index.js

In the server's implementation (`dist/index.js:1256`): 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: nst genResponse = await fetch(`https://openrouter.ai/api/v1/generation?id=${data.id}`, { headers: {

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/index.js

In the server's implementation (`dist/index.js:504`): A shell/process command assembled from concatenated or interpolated values is command injection when any part is attacker-influenced — the OWASP canonical RCE flow. Verify what reaches the interpolated value. This is read from the code itself — not from the tool description — so a poisoned server cannot hide it behind honest-looking metadata.

Evidence: tyCli(args2) { return execSync(`security ${args2.map(shellQuote).join(" ")} 2>/dev/null`, { encoding: "utf-8",

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/index.js

In a packaging/dev/install script (shipped, but not the server runtime) (`scripts/build-eval-dataset.ts: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: ecFileSync } from "node:child_process"; import { mkdirSync, writeFileSync } from "node:fs"; import { basename, dirname,

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 scripts/build-eval-dataset.ts

"mnemex" executes postinstall script(s) at install time. An install hook runs at install time; most are routine build/setup, but review what it does before trusting it.

Evidence: node scripts/postinstall.cjs || true

Fix: Review the scripts; install with --ignore-scripts where possible and vet what they do.

Location: package mnemex