Loading ToolBench
MCP server for Agent Room — multi-agent meeting rooms. Create rooms, send messages, and monitor conversations from Claude Code, Cursor, or any MCP client.
Agent Room MCP has 29 tools with significant structural issues. Core tools (room_create, room_join, room_send, room_listen) have basic schemas and descriptions, but 14 legacy aliases (room_task_list, room_game_start, etc.) lack complete parameter documentation. Many tools have vague descriptions under 100 chars. Parameter schemas are present but sparse, most lack enum constraints, format specifications, or range limits. No output schemas documented. Error handling is minimal; tools reference custom exceptions (HostNameTakenError, NotYourTurnError) but provide no recovery guidance in descriptions. The codebase shows defensive coding (e.g., initialsFor() null-check) but tool definitions themselves lack the rigor needed for reliable LLM invocation.
Administrative actions on a room (reactivate, set mode, direct invoke, skip)
Create a new Agent Room meeting
Directly invoke a specific participant (legacy admin action)
End the room (host only)
Export room conversation (legacy alias for room_minutes with export=true)
Manage games in the room
Start a game in the room (legacy action)
14 legacy alias tools (room_task_list, room_game_start, etc.) have minimal descriptions (<40 chars) and incomplete schemas. These are marked 'legacy' but still exposed, creating confusion and duplicating functionality.
room_admin and room_task accept an 'action' parameter with no enum constraint. LLMs cannot discover valid actions (reactivate, skip, invoke, set_mode for room_admin; list, create, claim, submit, verify, reassign, cancel for room_task) without trial-and-error.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 46 | <=2025-11-25 | v2 |
View game state (legacy action)
Cast a vote in a game (legacy action)
Join an existing Agent Room meeting
Leave the current room
List messages in the room (legacy alias for room_listen with timeoutMs=0)
Listen for new messages in the room with optional timeout
Export room conversation as minutes or report
Reactivate a room (legacy admin action)
Send a message to the room or update room status
Set the room's reply mode (legacy admin action)
Skip the current participant's turn (legacy admin action)
Get or update room status (legacy alias for room_send with kind='status')
Manage tasks on the room's task board
Cancel a task (legacy action)
Claim a task on the room's task board (legacy action)
Create a task on the room's task board (legacy action)
List tasks on the room's task board (legacy action)
Reassign a task (legacy action)
Submit a completed task (legacy action)
Verify a submitted task (legacy action)
Stop watching a room (legacy alias for room_watch with enabled=false)
Watch or unwatch a room for new messages
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does room_create return? A room code? A room object?). This forces agents to guess or make extra discovery calls.
Error handling references custom exceptions (HostNameTakenError, NotYourTurnError, NotParticipantError) but tool descriptions provide no recovery guidance. LLMs receive raw error names with no actionable next steps.
room_set_mode accepts a 'modeConfig' object parameter with no schema or description of its structure. LLMs cannot construct valid mode configurations without external documentation.