Nitpick MCP Server

Tools that read sensitive data ([export_data]) and tools that can send data out ([evaluate, upload_file]) are exposed together. An agent can move private data to the sink.

Evidence: sources [export_data] → sinks [evaluate, upload_file]

Fix: Keep secret-reading and egress capabilities on separate, separately-approved servers.

Location: flow export_data → evaluate

Tool "evaluate" appears to run shell commands or evaluate code (parameter "script"). Arbitrary execution driven by model input is one of the most dangerous MCP capabilities; combined with any untrusted input it becomes RCE.

Fix: Sandbox execution, allowlist commands/arguments, and never pass model output to a shell unescaped.

Location: tool evaluate

In the server's implementation (`dist/report.js:2`): 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"; const SEV_COLOR = { high: "#b91c1c", medium: "#c2410c", low: "#15803d" }; 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/report.js

In the server's implementation (`dist/util/sh.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: { execFile } from "node:child_process"; const MAX_BUFFER = 64 * 1024 * 1024; // 64MB — screenshots/hierarchy dumps are l

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/util/sh.js

Tool "evaluate" takes a command-shaped parameter "script" with no enum/pattern constraint. Free-form, model- or attacker-controlled arguments reaching a shell is the command-injection precondition.

Fix: Constrain the parameter (enum/pattern), or build the command from a fixed template with escaped args.

Location: tool evaluate · inputSchema.properties.script

In the server's implementation (`dist/doctor.js:28`): 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: st { chromium } = await import(spec); const p = chromium.executablePath(); return !!p && existsSync(p);

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/doctor.js

In the server's implementation (`dist/drivers/playwright.js:42`): 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: ({ chromium } = await import(spec)); } catch { throw new Error("Playwright is not installe

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/drivers/playwright.js

Tool "evaluate" can mutate/egress but declares no destructiveHint. Clients that don't default to spec-safe behavior may not prompt before running it.

Fix: Declare accurate annotations, and gate destructive tools on user confirmation regardless.

Location: tool evaluate

Tool "export_data" takes a path parameter "filename" with no constraint. Without a canonicalize-and-contain check (not visible statically), this permits ../ traversal outside the intended root.

Fix: Resolve and verify the path stays within an allowed root; reject traversal sequences.

Location: tool export_data · inputSchema.properties.filename