Loading ToolBench

Reading the current corpus…

mcp-windows-remote-shell MCP server · C · ToolBench

← Browse

mcp-windows-remote-shell

github.com/laatortuejaune/mcp-windows-remote-shell

MCP server (stdio) exposing run_command / check_agent over an outbound relay tunnel.

2026-07-28+STDIO3 tools

61/100

Definition Quality79

Protocol Readiness50

Supportability38

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Three tools with clear security-focused descriptions and proper input schemas. Tool names are action-oriented (current_target, check_agent, run_command). Descriptions are substantive (100-150 chars) and explain prerequisites and security constraints. However, parameter descriptions are minimal or missing for some inputs, and output schemas are not documented. The run_command tool has well-defined numeric constraints (timeout 1-3600s, command length 1-65536 chars) but lacks guidance on error recovery or output format. No tool annotations (readOnlyHint/destructiveHint) despite clear risk profiles.

Tools (3)

check_agentread onlyauthsource verified77/100

Check relay agent. Discover and check the relay agent automatically. The session must match the package prepared on the operator workstation; no manual pairing is needed.

current_targetread onlyauthsource verified77/100

Show relay target. Discover the target automatically using the persistent package credential. Refuse an unexpected or replaced session. The reported host name is untrusted. Does not contact the target.

run_commandwriteauthsource verified82/100

Run command on verified relay target. Automatically discover and authenticate the target from the prepared package, then run one shell command (cmd /c on Windows, sh -c on Unix). An unexpected or replaced session is refused. Each call is independent.

Issues (5)

HIGHtool ↗

Output schemas not documented. Tools return text responses but LLMs cannot infer structure, field names, or downstream chaining requirements.

current_targetcheck_agentrun_command

MEDIUMtool-annotations ↗

Missing tool annotations. run_command is destructive (WRITE risk) and check_agent is read-only, but no destructiveHint or readOnlyHint declared in schema.

run_commandcheck_agentcurrent_target

HIGHrecovery-guide ↗

Error handling lacks recovery guidance. RelayError messages are generic ('relay request failed'); LLMs cannot determine if error is retryable, user-fixable, or fatal.

run_commandcheck_agentcurrent_target

HIGHresponse-shaper ↗

run_command output format undefined. Agents cannot parse command exit codes, stdout, stderr, or timeout behavior from response.

run_command

Recommendations

Document output schema for each tool. E.g., run_command should return {exit_code: number, stdout: string, stderr: string, timed_out: boolean}.
Add tool annotations: readOnlyHint=true for current_target and check_agent; destructiveHint=true for run_command.
Enhance error messages with recovery guidance. E.g., 'Session mismatch: prepared session expired. Call check_agent() to refresh.' instead of generic 'relay request failed'.
Clarify idempotency and retry semantics for run_command. State whether repeated calls with same command are safe or if side effects accumulate.
Add parameter descriptions to inputSchema. E.g., timeout_seconds: 'Execution timeout in seconds (1-3600); defaults to RELAY_AGENT_TIMEOUT_S env var or 30s if unset.'

Score history

Overall score trend

First recorded score · v2 rubric

61/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	C	61	2026-07-28+	v2