An MCP server for sequential thinking with session management, memory banking, architecture decision recording, and package exploration capabilities
This server has 12 tools with complete JSON Schema definitions and mostly adequate descriptions. Many tools lack context about when to use them, prerequisites, or recovery guidance. Error handling is generic ('return {"error": str(e)}') with no actionable recovery hints. Parameter descriptions are present but terse. The server does not implement tool annotations (readOnlyHint/destructiveHint/idempotentHint), which would clarify intent for agents. Overall: functional schemas and adequate naming, but descriptions and error guidance fall short of production baselines.
Add a new thought to the current session with optional package exploration
Analyze current session completeness and insights
Create a new branch for alternative reasoning
Explore packages for given task
Export session or memories to file
List all saved sessions
Resume a specific session
Tool descriptions consistently under 50 chars; insufficient for LLM selection. E.g., 'List all saved sessions' (23 chars) lacks context about when to call it or what structure it reveals.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot determine which tools are safe to retry or have irreversible side effects. Tools like 'export_session', 'store_memory', and 'merge_branch' modify state but are not marked as destructive.
Error handling is uniformly generic. All tools catch exceptions and return {'error': str(e)}, which provides no recovery guidance. E.g., 'NoActiveSessionError' should suggest 'Call start_session() first.' instead of bare error message.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | <=2025-11-25 | v2 |
Merge a branch back into main reasoning
Search memories by tags or content with advanced filtering
Record an architecture decision with context
Start a new session (thinking, coding, or memory-focused)
Store a memory with code snippets
Parameter descriptions are terse (average ~15-20 chars) and omit format/constraint guidance. E.g., 'confidence' param described as 'Confidence level of the thought' but lacks range (0.0 - 1.0 assumed but not stated).
No output schema documented for any tool. LLMs cannot plan downstream calls or extract required data without knowing response structure.
query_memories accepts 'tags' and 'content_contains' with no input validation or clear error messages for invalid regex. Current code passes to session_manager with no visible sanitization against injection.