MCP server for managing and spawning Claude Code subagents
Server has 7 well-named tools with complete input schemas and descriptions. Naming follows verb_noun convention (list_, spawn_, create_, get_, cancel_). Descriptions are substantive (100-250 chars) and explain WHEN to use each tool. However, output schemas are not documented, LLMs cannot predict response structure. Error handling returns text-only responses without recovery guidance or error classification. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Parameters lack constraints (enums, ranges, patterns), e.g., spawn_subagent accepts arbitrary model strings and timeout values without validation hints.
Cancel a running subagent task by taskId.
Dynamically create a temporary subagent for a one-time task. The subagent definition exists only in memory and will be lost on server restart. After creation, use spawn_subagent with the specified type to execute tasks using this subagent.
Get the status and result of a previously spawned subagent task by taskId.
List all subagent tasks (running and completed) with their status.
List all available subagent types and their descriptions. Use this to discover which subagents can be spawned and what each one does.
Reload the subagent configuration from disk. Use this after modifying subagents.json to make changes take effect without restarting the MCP server. Returns the new list of available subagents.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Handlers return JSON text without schema declaration.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). spawn_subagent and create_subagent are WRITE operations; cancel_subagent is REVERSIBLE. LLMs cannot distinguish safe from risky tools.
Error responses lack recovery guidance. Handlers return plain text errors without actionable next steps or error classification (retryable vs user-fixable vs fatal).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Spawn a subagent to execute a task. The subagent runs as an independent Claude Code process with its own model, instructions, and environment configuration as defined in subagents.json. Use list_subagents first to see available types. The subagent will execute the task and return the result.
Parameter constraints missing. spawn_subagent.timeout and spawn_subagent.maxTokens accept unbounded numbers; model accepts arbitrary strings. No enums, ranges, or patterns to guide LLM input.
No pagination support. list_subagents and list_subagent_tasks lack limit/offset/cursor parameters. Large result sets could exhaust context window.