Loading ToolBench

Reading the current corpus…

claude-ssh-server MCP server · D · ToolBench

claude-ssh-server

github.com/jasondsmith72/claude-ssh-server

MCP server enabling direct SSH connections with IP/username/password - no config required. Fork of ssh-mcp with ad-hoc connection support.

2026-07-28+STDIO2 tools

51/100

Definition Quality58

Protocol Readiness50

Supportability41

Spec posture

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

Static source inference · medium confidence

Current-spec patterns detected

Summary

Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Both tools accept credentials as parameters (critical security issue), lack output schema documentation, and have minimal error guidance. Naming is verb-based but descriptions are functional rather than LLM-optimized. No tool annotations or composition patterns evident.

Tools (2)

direct_execwriteauthsource verified57/100

Execute a shell command on any SSH server by providing connection details directly. This tool allows ad-hoc connections without pre-configuration.

execwriteauthsource verified58/100

Execute a shell command on the remote SSH server and return the output. Uses connection details from CLI arguments.

Issues (5)

CRITICALsecret-injection ↗

Credentials exposed as tool parameters: password and privateKey are passed directly as parameters in direct_exec, violating secret-injection pattern. These will appear in agent logs and traces.

direct_exec

HIGHtool ↗

No output schema documented. Tools return results from execSshCommand but the response structure (success/failure, stdout/stderr, exit code) is not specified. LLMs cannot plan downstream actions without knowing return fields.

execdirect_exec

HIGHtool-description ↗

Missing parameter descriptions for critical fields: 'port' has a description but 'timeout' description is minimal. 'privateKey' should clarify format (PEM, OpenSSH, etc.). Parameter descriptions average 30-40 chars; baseline is 72 chars.

direct_exec

HIGHrecovery-guide ↗

Error handling lacks recovery guidance. Code throws McpError with generic messages like 'Unexpected error' without actionable next steps. LLMs cannot self-correct or retry intelligently.

Recommendations

Move password and privateKey to server-side configuration (environment variables or vault). Accept only a connection_profile_name parameter instead. This prevents credential leakage in logs.
Document output schema explicitly: return {success: boolean, stdout: string, stderr: string, exitCode: number, executedCommand: string}. Include this in tool descriptions and code comments.
Expand parameter descriptions to 50-100 chars each. For privateKey, specify: 'SSH private key content in PEM format (-----BEGIN PRIVATE KEY-----...). If using key authentication, omit password.'
Add actionable error messages: 'Authentication failed. Verify username, password, and host are correct. Try connecting manually: ssh -v user@host' instead of generic 'Unexpected error'.
Implement a dry_run parameter (boolean, default false) for both tools. When true, validate the command syntax and SSH connection but do not execute. This lets agents preview before committing.

Score history

Overall score trend

First recorded score · v2 rubric

51/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	51	2026-07-28+	v2