Kiro inspired Spec-Driven Development MCP Server
This server has significant gaps in definition quality. Both tools lack actionable descriptions that guide LLM tool selection, parameter descriptions are missing or minimal, and schemas are under-specified. The 'userInput' tool is problematic because it appears to delegate responsibility back to the LLM to call itself, a circular pattern that violates tool composition principles. The 'create-project-folders' tool has an empty schema and a description that delegates filesystem operations to the user rather than performing them. Output schemas are not documented. Naming is verb-first but descriptions fall short of the 10-1024 character baseline and lack context about when/why to use each tool.
Create the .project/ and .project/steering/ directory structure
Get user feedback and input during the spec creation process
create-project-folders description is misleading: says it 'creates' directories but actually asks the user to create them. Tool delegates its responsibility to the LLM/user rather than performing the stated action. Violates the command-tool pattern.
userInput tool creates a circular pattern: it asks the LLM to use a 'userInput' tool to get user input, but the tool itself is userInput. This is self-referential and violates tool composition. The tool should either directly wait for user input (blocking call to stdin/API) or return a structured multi-round-trip result with 'input_required' to signal the client to prompt the user.
Both tool descriptions are under 50 characters and lack actionable context. Baseline from production tools: 194 characters average. Current descriptions do not explain WHEN to use the tool, WHAT the preconditions are, or WHAT happens to the system state. LLMs cannot make informed selection decisions with such sparse descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 26 | - | v1 |
create-project-folders has an empty input schema (z.object({})) and no documented output schema. Tools must declare what inputs they accept and what outputs they return. The response text is returned but its structure is not formally documented.
userInput 'reason' parameter description is vague: 'The reason for asking user input' duplicates the parameter name without clarifying why this enum matters. Does it change behavior? Is it for logging? LLMs need context to make the decision to include it.
No error handling guidance. Neither tool documents what errors can occur, how to recover, or whether failures are retryable. If create-project-folders fails because directories exist, should the LLM retry or continue? Unclear.
Output schemas are not documented. The response.content structure is visible in code but not formally declared in a schema that the LLM can reason about. LLMs need to know what fields to extract from the response.
The server design conflates tool responsibilities with agent responsibilities. create-project-folders should actually create directories (or return an error); userInput should not ask the LLM to call itself. Tools should be self-contained and not delegate back to the agent.