Loading ToolBench
Cross-platform local MCP server and CLI for bitemporal long-term memory for AI agents
Mnemo has well-structured tool definitions with clear names and descriptions. All 7 tools are explicitly registered with @mcp_app.tool decorators and include descriptions (avg ~120 chars). Input schemas are present with typed parameters. However, parameter descriptions are sparse or missing for several tools (e.g., mnemo_search's 'limit' and 'min_tier' lack detail on constraints), and output schemas are not documented. Error handling returns plain text rather than structured guidance. The TOON serialization format is mentioned but not explained to LLMs. STDIO transport caps protocol readiness at 50, limiting overall score.
Detect and return knowledge / architectural debt: decaying working-tier facts, stale core facts, orphaned entities.
Inspect a single fact by ID, returned in TOON format.
Soft-delete a fact by closing its valid_end timestamp.
Reinforce a fact's salience by incrementing its access count.
Store a new memory fact. The AUDN classifier automatically detects duplicates (NOOP), updates, or fresh additions.
Perform an incremental AST code scan of the current repository to extract modules, classes, and dependencies into the knowledge graph.
Hybrid 3-channel search (Vector + FTS5 + Graph) with RRF fusion. Returns TOON-serialised results for token efficiency.
Output schemas not documented. Tools return TOON-serialized strings, but LLMs cannot parse the format without explicit schema documentation. mnemo_search returns 'sc:0.123456|ch:vector|f:(...)' but the structure is opaque to agents.
Parameter constraints missing. mnemo_search accepts 'limit' (default 10) and 'min_tier' (default 'archived') but descriptions lack min/max bounds, valid enum values, or format guidance. LLMs may pass invalid tier names or unreasonable limits.
Error messages are plain text without recovery guidance. mnemo_remember returns '[ERROR] invalid force_op...' but does not suggest what to do next. mnemo_search returns '[EMPTY] no matching memories' without suggesting alternative queries or filters.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
mnemo_get_debt_ledger has empty input schema ({}). Description does not explain what 'debt' means (decaying facts? orphaned entities?) or what structure is returned. LLMs cannot determine when to call this tool.
STDIO transport only. Server is not remotely accessible and cannot be used by hosted MCP clients. Limits deployment to local CLI use only.