Loading ToolBench
MCP server for GenSpec workflow - converts USER-STORIES.md into README, ROADMAP, and SYSTEM-ARCHITECTURE documents
GenSpec MCP has significant structural issues that prevent higher scores. All 4 tools share a critical flaw: they have NO input parameter descriptions whatsoever. The schemas show structure (e.g., start_genspec accepts userStory and userStoryUri with URI format), but the descriptions that accompany each parameter are completely absent. Additionally, 3 of 4 tools (generate_readme, generate_roadmap, generate_architecture) accept ZERO parameters despite performing state-changing operations. Tool descriptions themselves are present but generic/vague (e.g., 'Generate README, then continue through ROADMAP → SYSTEM-ARCHITECTURE' does not explain WHEN to use it, what the consequences are, or how it differs from start_genspec). Output schemas are not documented anywhere in the provided source, we cannot verify what fields are returned or how to chain these tools together. Error handling in the code shows generic Error() throws with no recovery guidance. Per the hard rules: parameters lack descriptions (cannot exceed 0-20), and output schemas are not visible (must score 0 for schema). This server would not pass production code review.
Generate only SYSTEM-ARCHITECTURE
Generate README, then continue through ROADMAP → SYSTEM-ARCHITECTURE
Generate ROADMAP, then continue through SYSTEM-ARCHITECTURE
Start full GenSpec workflow: README → ROADMAP → SYSTEM-ARCHITECTURE
MISSING: All parameter descriptions. start_genspec defines userStory and userStoryUri parameters but provides ZERO description text for either.
ZERO parameters on 3 stateful tools. generate_readme, generate_roadmap, generate_architecture accept NO input parameters yet perform file-writing operations. How does the LLM specify which user story to process? Which workspace? This forces magic behavior (e.g., assumes local USER-STORIES.md exists) and violates pattern:tool (tools should accept relevant inputs).
MISSING: Output schema documentation. Source code does not document what fields ToolOutputSchema returns (is it { phase, nextAction, draftPath }? Does it include file contents? Errors?). LLMs need to know what fields to expect.' No output documentation = LLMs cannot chain these tools or plan downstream steps.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 31 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 6 | - | v1 |
VAGUE tool descriptions. 'Generate README, then continue through ROADMAP → SYSTEM-ARCHITECTURE' is generic and does not explain: (1) what content goes into README vs downstream docs, (2) when to call generate_readme vs start_genspec, (3) whether it modifies files on disk or returns drafts, (4) what 'nextAction: approve' means.
NO error recovery guidance. Code throws generic Error() with messages like 'Phase execution failed' but provides no hint to the LLM about next steps (retry? different tool? inspect logs?). Per pattern:recovery-guide: 'Error responses must tell the LLM what to do next.' Current approach leaves agent stuck.
Composition issue: start_genspec and generate_readme both start a README phase. Difference is unclear. start_genspec accepts userStory/userStoryUri, but generate_readme reads from local filesystem. No description explains when to call which.
Missing idempotency documentation. If an LLM retries generate_readme, will it overwrite the previous README or append? Will it create duplicate entries? No description or code comments clarify side effects. Per pattern:idempotent-operation: 'Make tools produce the same result on repeated calls.' Not addressed.
Missing natural identifiers. userStoryUri parameter expects a URI, but no description clarifies whether it accepts file:// URIs, HTTP URLs, or something else.