MCP server for queuing live speech playback with stored SpeakSwiftly voice profiles
Single tool 'generate_speech' has a complete input schema with proper JSON Schema structure and nested objects (requestContextInputSchema, generationLocationInputSchema). Description is detailed (250+ chars) and explains purpose, parameters, and context. However, description is overly technical and verbose for LLM parsing; parameter descriptions are missing entirely from the schema (only type info visible); output schema is completely undocumented; no error handling guidance provided. The tool performs a state-changing operation (queuing speech) but lacks confirmation/dry-run pattern. Schema is visible and well-formed, preventing a lower score, but missing parameter descriptions and output documentation are significant gaps.
Queue live speech playback with a stored SpeakSwiftly voice profile. Use this when the user wants audible output now. The server applies live-speech purpose plus MCP client and tool provenance by default; optionally provide profile_name to override the server's configured default voice profile plus text_profile_id, request_context, cwd, and repo_root when the input needs richer caller metadata. request_context.prefacePolicy may be default, always, or never.
Parameter descriptions missing from schema. Schema defines 'text', 'profile_name', 'text_profile_id', 'cwd', 'repo_root', 'qwen_pre_model_text_chunking', 'generation_location' but only 'text' has a description in the input schema object. LLMs cannot infer what 'qwen_pre_model_text_chunking' or 'generation_location' control without explicit descriptions.
Output schema completely undocumented. Tool returns a CallTool.Result but no documentation of response structure, fields, or format. LLMs cannot plan downstream actions or extract relevant data without knowing what the response contains.
No error handling guidance. Tool description does not explain failure modes, recovery steps, or what errors the LLM should expect. Handler code shows validation (missing voice profile) but error messages are not documented in the tool definition.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
State-changing operation lacks confirmation pattern. Tool queues live speech playback (irreversible side effect) but provides no dry-run, preview, or confirmation step. Agents cannot safely test before committing.
Tool description is overly technical and verbose (250+ chars). Mentions 'live-speech purpose', 'MCP client and tool provenance', 'prefacePolicy' enum values inline. LLM-optimized descriptions should be 50-200 chars and focus on WHEN to use the tool, not implementation details.