Loading ToolBench
MCP server exposing Mneia to Claude Code, Cursor, Codex, and any MCP client. Mneia is the shared project memory for repositories, providing checkpoint, rehydrate, and handoff operations for AI agents.
Critical failures across all 12 tools. Descriptions are present but generic (10-50 chars), below the 50-200 char LLM-optimized baseline. No parameter descriptions exist, the schema objects have empty properties with no type or description fields. Tool names use inconsistent patterns (mneia_* prefix is non-standard; verb_noun convention not followed). No output schemas documented. No error handling guidance. This server is non-functional for agent use.
Records a decision, constraint, or open question so the next session inherits it.
Captures the session as a whole at a task or day boundary.
Creates a handoff to transfer work to another actor.
Lists handoffs that were handed to the current actor.
Receives a handoff that was created for the current actor.
Returns the active constraints, the decisions already made, and the open questions for the current task.
Retires a context item from the repository memory.
All 12 tools have empty input schemas (properties: {}). LLMs cannot infer parameter names, types, or constraints.
Tool descriptions are 10-50 characters (e.g., 'Returns the active constraints...' is 50 chars). Below 50-char minimum for LLM optimization. Baseline is 50-200 chars for clarity.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | 2025-06-18+ | v2 |
Confirms a decision on an item that was pending review.
Reads everything still waiting on a person to confirm. This is read-only.
Searches for context items in the repository memory.
Lists who has been in the repository before the current session.
Returns the names and ids of team members that mneia_handoff_create accepts as toActor.
Tool names use 'mneia_*' prefix instead of verb_noun convention (e.g., 'mneia_rehydrate' should be 'get_context' or 'load_session'). LLMs cannot infer intent from non-standard names.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does mneia_rehydrate return? What fields are available?).
No error handling guidance. Tools marked with WRITE risk (assert, checkpoint, handoff_create, etc.) have no recovery instructions or confirmation patterns to prevent accidental data loss.