Multi-persona engineering mentor with 64 specialized AI personas orchestrating collaborative guidance, session memory, and git-aware context
Sensei MCP has 5 tools with verb-noun naming (get_, record_, validate_, list_) and descriptions present. However, critical gaps undermine quality: (1) Input schemas are inferred from function signatures but NOT explicitly visible in JSON Schema format in the code, cannot verify type definitions, constraints, or enums. (2) Parameter descriptions exist in docstrings but are generic ('Session identifier', 'Absolute path') without format/constraint guidance. (3) Output schemas are completely undocumented, tools return markdown strings or confirmation messages with no structured schema declaration. (4) No error handling guidance; tools silently fail or return generic strings. (5) session_id and project_root parameters appear in every tool, suggesting stateful session management (deprecated pattern). Per-tool average: 52.
Get relevant Sensei engineering context for the current task.
Get a summary of the current session's decisions and context.
List all available sessions in the global directory. (Note: Does not list local project sessions)
Record an architectural or technical decision for this session.
Validate code or design against Sensei engineering standards.
No explicit JSON Schema definitions visible in source code. Input schemas inferred from function signatures only; cannot verify type constraints, enums, min/max, or patterns. Violates pattern:constrained-input and pattern:tool-description.
Output schemas completely undocumented. All tools return markdown strings or plain text confirmations with no structured schema declaration. LLMs cannot parse or chain results. Violates pattern:response-shaper.
Parameter descriptions are generic and lack format/constraint guidance. 'Session identifier' does not explain format, length, or allowed characters. 'Absolute path to the project root' lacks validation rules. Violates pattern:tool-description.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | 2026-07-28+ | v2 |
No error handling guidance. Tools return generic strings on failure with no recovery hints. LLMs cannot determine if errors are retryable, user-fixable, or fatal. Violates pattern:recovery-guide and pattern:error-classification.
Stateful session management via session_id and project_root parameters in every tool suggests reliance on deprecated Mcp-Session-Id pattern. Protocol should be stateless; session context should be passed via tool parameters or configuration, not maintained server-side.