This MCP server defines 4 tools with complete input schemas and descriptions. Tool naming is strong (verb-first), and descriptions clearly state what each tool does. However, output schemas are not documented, the rubric requires documentation of return types for A+ rating, and this is a notable gap. Error handling exists but is minimal (bare 'not found' messages). Parameter descriptions are present but some lack detail about constraints (e.g., 'limit' has no min/max bounds documented). The semantic search tool is well-designed with enum constraints on chunkType, but composition could be tighter, search_sessions and get_session_context together form a common pattern that could be unified. Overall, this is above-average for community MCP servers but falls short of 'A' grade due to missing output schema documentation and limited error guidance.
Get full context for a specific session — summary, timeline, key decisions and code changes.
Generate a short summary of a session — what was done, which files were changed, key decisions.
List all documented decisions across sessions for a project.
Semantic search across all Claude Code session transcripts. Returns relevant chunks with context.
Output schemas are not documented. The rubric mandates documented return types for A-grade tools. Callers (LLMs) cannot predict what fields will be present in responses, forcing them to guess field names when chaining tools.
Error messages lack recovery guidance. When a session is 'not found', the tool returns a bare string rather than suggesting next steps (e.g., 'Try search_sessions() first'). This violates the recovery-guide pattern.
Numeric parameters (limit, page_size) lack min/max bounds in descriptions. The 'limit' param in search_sessions defaults to 5 and list_decisions to 20, but no upper bounds are stated. Per the rubric, unbounded numbers let LLMs pass absurd values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 11 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 47 | - | v1 |
search_sessions parameter descriptions could be more explicit about format. The 'query' param says 'Search query in natural language' but does not specify length limits or character restrictions. Per the rubric, constraints should be in descriptions, not just JSON Schema.
Tools return plain text (markdown formatted strings) rather than structured JSON objects. This forces LLMs to parse unstructured text, which is error-prone and wastes tokens. The pattern:response-shaper mandates returning structured objects with typed fields.