Loading ToolBench
Persistent memory for AI coding agents: an MCP server that records your sessions, normalizes them into a searchable knowledge graph, and hands the context back at the start of the next one.
simple-rick defines 8 tools with complete input schemas and descriptions. Naming follows verb_noun convention (simple_rick_*). Descriptions are present but generic (avg 95 chars, below 194-char baseline). Schemas are well-formed with types and required fields. However, output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling is absent: no recovery guidance, no categorization of retryable vs fatal errors. Parameters lack format constraints (e.g., 'limit' is unbounded number). No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Composition is sound, each tool has one responsibility, but missing idempotency guarantees and confirmation patterns for destructive ops (simple_rick_close, simple_rick_decision, simple_rick_link).
Ask a question about the code, past decisions or how things connect.
Session start: returns context, open issues, learnings and recommendations. Call this at the beginning of every session.
Session end: normalizes all message pairs, extracts learnings and creates embeddings. Always call this before ending a session.
Explicitly record an architecture or design decision.
Initialize Simple Rick for a new project. Captures project name, stack and structure, and builds the initial context.
Generate insights from the knowledge base. Finds correlations, trends and anomalies between entities and validates them with an LLM.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination) for downstream chaining or data extraction.
No error handling guidance. Tools lack recovery hints (e.g., 'If search fails, try simple_rick_ask instead'). No error categorization (retryable vs fatal).
Destructive tools lack confirmation/dry-run pattern. simple_rick_close and simple_rick_decision modify state irreversibly without confirmation step.
Tool annotations missing. No readOnlyHint/destructiveHint/idempotentHint despite clear risk levels (WRITE vs READ_ONLY marked in metadata).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Create an explicit cross-link between two chunks or concepts.
Semantic search across the entire project history.
Parameter constraints missing. 'limit' in simple_rick_search is unbounded number; 'mode' in simple_rick_insights has enum but no description of each mode's behavior.