Remote Ssh (shiharu) MCP Server

In the server's implementation (`dist/src/shared/config.js:21`): 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: privateKey: "~/.ssh/id_rsa", }, }; // ─── Cached config directory path (resolved once on first call) ───────────

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/src/shared/config.js

Tool "exec_session_start" appears to run shell commands or evaluate code (keyword "exec" in tool name). 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 exec_session_start

Tool "exec" appears to run shell commands or evaluate code (keyword "exec" in tool name, 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 exec

Tool "exec_session_close" appears to run shell commands or evaluate code (keyword "exec" in tool name). 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 exec_session_close

In the server's implementation (`dist/src/gateway/transport.js:102`): 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: sion lifecycle: start → exec (multiple) → close. " + "Returns session_id and initial working directory.",

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/src/gateway/transport.js

In the server's implementation (`dist/src/managers/exec.js:159`): 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: pdated cwd. */ exec(sessionId, command, opts) { const session = this.sessions.get(sessionId); i

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/src/managers/exec.js

Untrusted-input tools ([sftp_download]) co-exist with external-action tools ([exec_session_start, exec, exec_session_close]). A prompt injection could cause unwanted external actions, though no direct sensitive-data leak path was found.

Evidence: untrusted [sftp_download] → sinks [exec_session_start, exec, exec_session_close]

Fix: Require confirmation for state-changing/egress actions triggered after processing untrusted content.

Location: flow sftp_download → exec_session_start

Tool "sftp_download" takes a URL/host parameter "host" with no allowlist/pattern. An outbound-request tool with an unbounded destination enables SSRF and cloud-metadata access (e.g. 169.254.169.254).

Fix: Allowlist destinations or constrain the parameter; block private/link-local addresses server-side.

Location: tool sftp_download · inputSchema.properties.host

Tool "exec" 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 exec · inputSchema.properties.command

Tool "exec_session_start" 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 exec_session_start

Tool "exec" 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 exec

Tool "exec_session_close" 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 exec_session_close