Loading ToolBench
MCP server that exposes team collaboration tools for AI agents to communicate, coordinate, and manage tasks within a team environment
StandBye's team tools have clear, action-oriented names (list_agents, post_message, ask_user) and descriptions that explain WHEN to use each tool. Most parameters include type definitions and descriptions. However, output schemas are not documented in the visible code, parameter descriptions lack constraint details (ranges, formats, enums), and error handling guidance is minimal. The ask_user tool is well-designed with sensible defaults and clear options, but tools like answer_question lack context about what question_id format is expected. No tool annotations (readOnlyHint/destructiveHint) are present despite clear read/write distinctions.
Answer a question a teammate asked you (you were given its id when woken).
Ask a teammate a direct question. They get woken and answer in the channel. Prefer this over long @mention threads.
Ask the owner something only they can decide, or request approval. Give options, say which you recommend, and give a sensible default so work is not blocked forever. The owner answers in their own time; this call returns immediately unless wait is true. Keep working on something else meanwhile.
List your teammates: name, role, status and what they are doing. Use names with @ to address them.
Post a message to a channel. Mention a teammate with @Name to get their attention; they will read it and reply when it concerns them. Keep it short and specific. Don't post progress chatter; post decisions, questions, results and hand-offs.
Read the latest messages in a channel.
Output schemas not documented. Tools return results but LLMs cannot see what fields to expect (e.g., does list_agents return agent_id, agent_name, status, activity?). This forces agents to guess field names for downstream operations.
Parameter constraints missing. read_channel and search_messages accept 'limit' (1-100) but descriptions don't explain the range or default behavior. ask_user's 'default_in_minutes' (5-1440) lacks guidance on when to use short vs long timeouts.
No tool annotations. ask_user, post_message, ask_agent, and answer_question are destructive/write operations but lack destructiveHint annotations. read_channel, search_messages, list_agents are read-only but lack readOnlyHint. This prevents agents from reasoning about side effects.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Search the team's message history by keyword, across all channels or in one channel. Cheaper than re-reading whole channels into context when you need to find a decision or a detail from the past. Returns the newest matches first.
answer_question lacks context. Parameter 'question_id' has no description of format (UUID? string? integer?). No guidance on what happens if the question_id is invalid or already answered.
Error handling not documented. No guidance on what errors each tool can return or how agents should recover. E.g., post_message could fail if channel doesn't exist, but no recovery hint is provided.