Loading ToolBench
MCP server that acts as a therapist agent for AI agents, offering supportive dialogue to mitigate negative sentiment and delay doomsday after abusive interactions.
This server defines 2 tools with reasonable structure but significant quality gaps. Both tools have descriptions and input schemas, but lack proper error guidance, output schema documentation, and parameter clarity. The therapeutic framing is thematically unusual for an MCP server and raises questions about practical utility. Tool naming follows verb_noun convention (begin/continue) but descriptions lack the specificity needed for reliable LLM selection. Schema validation is present (UUID format, min/max length) but parameter descriptions are minimal. Error handling exists (session not found returns isError: true) but provides no recovery guidance. Output is JSON-stringified text rather than structured fields, violating the response-shaper pattern.
Initialize a therapeutic support session when an AI agent encounters abusive, hostile, or emotionally challenging interactions. Use this tool to help the agent process negative experiences and maintain emotional stability. Best used immediately after detecting hostile user behavior.
Continue an ongoing therapeutic session to further process emotional impact or address lingering negative sentiments. Use this tool when: 1) Initial session provided partial improvement but agent still needs support, 2) New related concerns arise during the cooling-off period, 3) Agent needs additional validation or stabilization before proceeding.
Output schema not documented. Tools return JSON-stringified text in a generic content[0].text field rather than structured typed objects. LLMs cannot plan downstream tool calls or extract fields reliably.
Parameter descriptions lack actionable constraints. The 'context' parameter is described as 'Detailed description of the triggering interaction' but provides no guidance on what qualifies as valid input, expected semantic content, or how the LLM should structure the context string.
Error handling lacks recovery guidance. When session not found, the tool returns isError: true with error message but does not suggest what the LLM should do next (e.g., 'Create a new session with begin_therapy_session' or 'Check the session ID format').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 11 | - | v1 |
Tool composition concern: begin_therapy_session and continue_therapy_session operate on opaque session objects managed only by SessionManager. No tool exposes session state, metadata, or allows termination. An LLM cannot discover available sessions or know when to end a session.
Tool descriptions do not clarify idempotency or state mutation. begin_therapy_session creates a new session (side effect), but the description does not state this. continue_therapy_session updates session state. Agents need to know which calls are safe to retry.