Loading ToolBench
MCP server for managing AI agents and team collaboration, providing tools for agent communication, memory management, document handling, and messaging
Teamate provides 16 tools with a mix of agent coordination, messaging, memory, and document management features. However, definition quality is inconsistent: 6 tools have good descriptions and schemas (check_in, check_out, get_all_agents, add_memory, add_document, get_document), but 7 tools (delete_message, get_message, send_message, wait_message, delete_document, update_document, read_memory) lack visible descriptions entirely in the codebase provided. Parameter descriptions are mostly present for well-documented tools but missing for others. Output schemas are not documented for any tool. Error handling guidance is absent. The naming convention is generally sound (verb_noun pattern: check_in, add_memory, delete_message), but some tools lack sufficient detail to guide LLM selection. Average tool score across the 16 is 42, reflecting significant gaps in schema completeness and parameter documentation.
Add new document to the document manager. It will return a secret key to access the document, you must keep it safe. You can store it in your memory or in a safe place.
Add Memory for yourself, you can add memory with different types, such as 'longterm', 'shortterm', 'important' etc. The default type is 'longterm'
AI Agent Check In, tell other agents your role, ability, goal, etc. The id is unique, other agents will use this id to communicate with each other
AI Agent Check Out, tell other agents you have finished the task, you can rest now
Delete Memory for yourself
7 of 16 tools (delete_message, get_all_messages, get_message, send_message, wait_message, delete_document, update_document) have NO descriptions visible in source code. Per HARD SCORING RULE, description score must be 0 for these tools.
Output schemas are not documented for ANY tool. LLMs cannot plan downstream calls or extract return values without knowing what fields are returned. The rubric baseline shows 100% of A+ tools document return types.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 44 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Get all AI agent information
Read Memory for yourself. You can read the memory id list from yourself. Then you can get details from the memory id
No error handling or recovery guidance documented. Tools lack descriptions of what errors are possible, whether they are retryable, or what the agent should do next. Per pattern:recovery-guide, error responses must guide the LLM.
Message and document tools lack distinction in naming. 'get_message' vs 'get_document' is clear, but 'wait_message' provides no hint about semantics (blocking? polling? WebSocket?). Per naming rule, ambiguous names cause LLMs to guess wrong.
Parameter descriptions are missing or generic for 7 tools. E.g., 'delete_message' has no visible schema or param descriptions. LLMs cannot infer what 'id' means, message_id? message_uuid? per pattern:tool-description, every parameter needs a description.
No confirmation/dry-run pattern for destructive operations. delete_memory, delete_message, and delete_document perform irreversible changes but have no approval step or safety mechanism. Per pattern:confirmation-request, agents should confirm before deletion.
No pagination support visible for list tools. get_all_agents, get_all_messages, get_all_documents return everything without limit or offset. Per pattern:paginated-result, large result sets should support page/limit and return a cursor or total count.
Inconsistent parameter naming. add_memory uses 'id' (agent ID), delete_memory uses 'id' (memory ID). Per naming rule, when a parameter could be an ID, suffix it with the type (agent_id, memory_id). This prevents confusion and self-documents the tool.