Phren Cli MCP Server

This server (without client built-ins) exposes a complete data-exfiltration chain: get_memory_detail → add_custom_hook → remove_custom_hook. Untrusted input is ingested, private data is read, and it can be sent to an external sink via the agent composing the tools (→). Static analysis proves the primitive exists, not that a specific run will occur.

Fix: Remove one leg of the trifecta: isolate untrusted-input tools from secret-reading tools and from egress tools, or require human approval between them.

Location: flow get_memory_detail → add_custom_hook → remove_custom_hook

Tool "add_custom_hook" can both read sensitive data and send data to an external destination. Even without an explicit untrusted-input leg, this is a single-call data-exfiltration path if the model is ever manipulated.

Fix: Separate reading from sending; require explicit user confirmation before egress of file/secret contents.

Location: flow add_custom_hook

In the server's implementation (`dist/generated/memory-ui-graph.browser.js:2`): 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: {let t=U_(i,e),{Body:n}=new Function(t)();return n}function U_(i,e){return` ${G_(i,e)} ${O_(i,e)} return {Body: Body, Ve

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/generated/memory-ui-graph.browser.js

In the server's implementation (`dist/memory-ui-graph.runtime.js:2`): 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: {let t=U_(i,e),{Body:n}=new Function(t)();return n}function U_(i,e){return` ${G_(i,e)} ${O_(i,e)} return {Body: Body, Ve

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/memory-ui-graph.runtime.js

Tool "add_custom_hook" appears to run shell commands or evaluate code (parameter "command"). 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 add_custom_hook

Tool "remove_custom_hook" appears to run shell commands or evaluate code (parameter "command"). 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 remove_custom_hook

In the server's implementation (`dist/generated/memory-ui-graph.browser.js:2`): 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: {let t=U_(i,e),{Body:n}=new Function(t)();return n}function U_(i,e){return` ${G_(i,e)} ${O_(i,e)} return {Body: Body, Ve

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/generated/memory-ui-graph.browser.js

In the server's implementation (`dist/memory-ui-graph.runtime.js:2`): 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: {let t=U_(i,e),{Body:n}=new Function(t)();return n}function U_(i,e){return` ${G_(i,e)} ${O_(i,e)} return {Body: Body, Ve

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/memory-ui-graph.runtime.js

In the server's implementation (`dist/agents/providers/herdr.js:11`): 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: { execFileSync } from "child_process"; import { clampInt, resolveExecCommand } from "../../utils-helpers.js"; import {

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/agents/providers/herdr.js

In the server's implementation (`dist/cli/extract.js:15`): 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: { execFileSync } from "child_process"; import { resolveRuntimeProfile } from "../runtime-profile.js"; import { buildCit

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/extract.js

In the server's implementation (`dist/cli/govern.js:156`): 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: eSync } = await import("child_process"); const report = { gitGcRan: false, commitsSquashed: 0,

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/govern.js

In the server's implementation (`dist/cli/namespaces-store.js:3`): 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: { execFileSync } from "child_process"; import { getPhrenPath } from "../shared.js"; import { isValidProjectName, errorM

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/namespaces-store.js

In the server's implementation (`dist/cli/ops.js:3`): 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: { execFileSync } from "child_process"; import { EXEC_TIMEOUT_MS, findProjectNameCaseInsensitive, getPhrenPath, normaliz

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/ops.js

In the server's implementation (`dist/cli/session-git.js:5`): 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: { execFileSync } from "child_process"; import * as fs from "fs"; import * as path from "path"; import { EXEC_TIMEOUT_MS

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/session-git.js

In the server's implementation (`dist/cli/team.js:7`): 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: { execFileSync } from "child_process"; import { getPhrenPath } from "../shared.js"; import { FINDINGS_FILENAME } from "

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/team.js

In the server's implementation (`dist/content/validate.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: { execFileSync } from "child_process"; import { debugLog, EXEC_TIMEOUT_MS, getProjectDirs } from "../shared.js"; import

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/content/validate.js

In the server's implementation (`dist/editor/launch.js:14`): 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: { execFileSync } from "child_process"; import { resolveExecCommand } from "../utils-helpers.js"; import { errorMessage

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/editor/launch.js

In the server's implementation (`dist/hooks.js:5`): 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: { execFileSync } from "child_process"; import { isIP } from "net"; import { fileURLToPath } from "url"; import { EXEC_T

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/hooks.js

In the server's implementation (`dist/embedding.js:178`): 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("https://api.openai.com/v1/embeddings", { method: "POST", 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/embedding.js

Tool "add_custom_hook" takes a command-shaped parameter "command" 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 add_custom_hook · inputSchema.properties.command

Tool "remove_custom_hook" takes a command-shaped parameter "command" 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 remove_custom_hook · inputSchema.properties.command

Tool "add_custom_hook" 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 add_custom_hook

Tool "remove_custom_hook" 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 remove_custom_hook