Loading ToolBench

Reading the current corpus…

2native-ssh-mcp MCP server · D · ToolBench

← Browse

2native-ssh-mcp

github.com/daidaiJ/2native-ssh-mcp

An SSH-based MCP server that exposes SSH command execution and file transfer as MCP tools over stdio or streamable HTTP

2026-07-28+4 tools

53/100

Definition Quality62

Protocol Readiness48

Supportability40

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

Four tools with clear verb-noun naming (list-servers, execute-command, file-transfer, session). All have descriptions (194 - 250 chars, within baseline 34 - 392). Input schemas present with typed parameters and descriptions. Tool annotations (readOnlyAnnotation) visible for list-servers. Error handling includes structured ToolError with code/message/retriable fields and partial result recovery. Main gaps: no output schema documentation in visible code; parameter descriptions are functional but terse (avg ~40 chars vs baseline 72); no enum constraints on 'action' parameter in session tool despite having fixed values; missing idempotent/destructive hints on most tools.

Tools (4)

execute-commanddestructive50/100

Execute a shell command on a configured SSH server, with optional approval gate for destructive commands

file-transferwrite50/100

Transfer files to/from a configured SSH server via SFTP

list-serversread onlysource verified83/100

List configured SSH servers (metadata, connection status, system info, last few recorded commands) and active named sessions. Call this first to pick connectionName or sessionName — the recent commands help resume context from a previous session.

sessionwrite50/100

Manage named SSH sessions for maintaining state across multiple commands

Issues (5)

HIGHtool ↗

Output schemas not documented in visible code. LLMs cannot plan downstream tool calls or extract required fields without knowing what list-servers, execute-command, file-transfer, and session return.

list-serversexecute-commandfile-transfersession

HIGHconstrained-input ↗

session tool 'action' parameter accepts enum values [create, list, close] but is defined as free-form string, not constrained enum. LLMs may hallucinate invalid actions.

session

MEDIUMtool-annotation ↗

execute-command lacks destructive hint annotation despite being marked DESTRUCTIVE risk. Tool annotations (readOnlyHint/destructiveHint) should be applied to guide agent behavior.

execute-command

MEDIUMtool-description ↗

Parameter descriptions are terse (avg ~40 chars vs baseline 72). 'connectionName' and 'sessionName' lack format guidance; 'command' lacks examples of safe vs destructive patterns; 'operation' lacks explanation of when to use upload vs download.

Recommendations

Add explicit output schema documentation for each tool. Example: 'Returns {servers: [{name, status, lastCommand, systemInfo}], sessions: [{name, connectionName, createdAt}]}'.
Convert session 'action' parameter to constrained enum in schema: {"type": "string", "enum": ["create", "list", "close"]}.
Apply destructiveHint annotation to execute-command and file-transfer (upload/download) tools to signal risk to agents.
Expand parameter descriptions to 60 - 100 chars with format guidance. E.g., 'connectionName: Name of configured SSH server (e.g., prod-db-01). Call list-servers first to see available connections.'
Document result limits and pagination in list-servers description: 'Returns up to 50 most recent servers and active sessions. For large inventories, implement cursor-based pagination.'

Score history

Overall score trend

First recorded score · v2 rubric

53/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	53	2026-07-28+	v2