Loading ToolBench

Reading the current corpus…

hermes-claude-code MCP server · D · ToolBench

← Browse

hermes-claude-code

github.com/DevvGwardo/hermes-claude-code

MCP server + Hermes plugin for delegating coding tasks to Claude Code CLI

2026-07-28+STDIO3 tools

56/100

Definition Quality73

Protocol Readiness50

Supportability31

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 complete schemas and descriptions, but significant gaps in parameter documentation and error handling guidance. Tool names follow verb_noun convention (delegate, research, batch). Descriptions are detailed (150-250 chars) and explain WHEN to use each tool. However, many optional parameters lack descriptions or have minimal guidance. No error recovery patterns documented. Output schemas are not formally specified. The server delegates to an external CLI (Claude Code), which limits error handling visibility.

Tools (3)

claude_code_batchwritesource verified68/100

Run multiple coding tasks in parallel using Claude Code CLI. Each task runs in its own isolated session. Tasks can be a mix of read-only research and autonomous coding work. Use this to parallelize independent sub-tasks: analyzing multiple files, writing multiple functions, running tests in parallel, etc. Results are returned in order with per-task timing.

claude_code_delegatewritesource verified77/100

Delegate a coding task to Claude Code CLI, an AI coding agent with full tool access. Claude Code will autonomously execute the task — reading/writing files, running terminal commands, searching code, making git commits — and return the result. Use this to offload implementation sub-tasks: writing functions, creating files, fixing bugs, refactoring, running tests, creating commits, etc. Claude Code runs with full permissions (bypassPermissions mode) and works in the specified directory. Best for self-contained tasks with clear instructions. Each task gets its own isolated session. IMPORTANT: Provide ALL context the sub-agent needs — it has no memory of your conversation. Include file paths, error messages, expected behavior, constraints, and any relevant code snippets.

claude_code_researchread onlysource verified73/100

Ask Claude Code to research or explore a codebase question WITHOUT making any changes. Claude Code will read files, search code, analyze architecture, and report findings. It will NOT write, edit, or delete any files. Use this for investigation tasks: understanding code, finding patterns, analyzing dependencies, reading documentation, tracing call chains, reviewing code quality.

Issues (5)

HIGHresponse-shaper ↗

Output schemas not documented. Tools return raw CLI output as text without specifying structure, field types, or pagination. LLMs cannot plan downstream calls or extract structured data reliably.

claude_code_delegateclaude_code_researchclaude_code_batch

HIGHtool-description ↗

Many optional parameters lack descriptions or have minimal guidance. 'system_prompt', 'session_id', 'max_budget_usd', 'allowed_tools', 'add_dirs' are present but descriptions are terse (10-30 chars). LLMs cannot infer when to use these parameters.

claude_code_delegateclaude_code_batch

HIGHrecovery-guide ↗

No error recovery guidance. When Claude Code CLI fails (exit code != 0), the tool returns raw stderr. No actionable error messages, no categorization (retryable vs fatal), no suggestions for next steps.

claude_code_delegateclaude_code_researchclaude_code_batch

Recommendations

Document output schemas for all three tools. Specify: success response structure (e.g. {text: string, sessionId?: string, exitCode: number}), error response structure, and pagination if applicable. Use JSON Schema format.
Expand parameter descriptions to 50-100 chars each. For 'system_prompt', add: 'Custom system prompt to prepend to Claude Code's default instructions. Use to set specific behavior, constraints, or persona. Max 2000 chars.' For 'allowed_tools', clarify: 'Restrict Claude Code to specific tools (e.g. ["Read", "Write", "Edit", "Bash", "Grep", "Glob"]). By default, all tools are available. Useful for limiting permissions.'
Implement error recovery guidance. Parse exit codes and stderr; return structured errors like: {error: 'timeout', message: 'Claude Code exceeded 10-minute timeout. Try breaking the task into smaller steps or increasing max_budget_usd.', retryable: true}. Map common CLI errors to actionable guidance.
Add per-task result structure to claude_code_batch. Return {tasks: [{task_index: 0, status: 'success'|'failure', result: string, error?: string, duration_ms: number}], total_duration_ms: number}. This enables agents to identify which tasks succeeded and which failed.
Add a dry_run parameter (boolean, optional, default false) to claude_code_delegate. When true, Claude Code returns a plan without executing. This enables confirmation workflows: agent calls with dry_run=true, shows plan to user, then calls with dry_run=false to execute.

Score history

Overall score trend

First recorded score · v2 rubric

56/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	56	2026-07-28+	v2