Loading ToolBench
MCP server exposing Mnemoscope's predict_rot, get_tiered_read, and journal tools to any MCP-compatible client.
Four tools with complete input schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. Tool names follow verb_noun convention (predict_rot, get_tiered_read, record_journal, read_journal). Descriptions range 100-200 chars, meeting baseline minimums. However, parameters lack descriptions in several cases (e.g., working_max_age_days, episodic_max_age_days in get_tiered_read have descriptions but content_before/content_after in record_journal lack clarity on when they're required). No documented output schemas visible in source. Error handling and recovery guidance absent. Security considerations for file path traversal not evident.
Returns a tiered (working / episodic / semantic) view of a Markdown vault, designed to be used as a compacted context substitute instead of injecting the full vault.
Compute a predictive context-rot risk score (0-100) for a Markdown vault before injecting it into an LLM. Returns the score, a breakdown of contributing factors, and the 5 highest-risk notes.
Read the journal entries for a vault. Optionally filter by session_id.
Append an entry to the local signed journal of agent operations on a vault. Use this to record reads, writes, creates, or deletes performed during a session.
Output schemas not documented. LLMs cannot plan downstream operations or extract required fields from predict_rot (score breakdown, top 5 notes) or get_tiered_read (tiered structure). Agents must guess response structure.
record_journal parameters content_before and content_after marked optional but lack descriptions explaining when they're required or how they're used. Ambiguous optional fields invite misuse.
vault_path parameter accepts absolute file paths with no validation guidance. No mention of path traversal protection, symlink handling, or permission checks. Security risk if agent is compromised.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
No error handling guidance. What happens if vault_path doesn't exist? If journal is corrupted? If signature verification fails? LLM receives no recovery hints.
record_journal accepts free-form session_id string with no format guidance. Should be UUID or documented pattern to prevent collisions and enable audit trail parsing.