Loading ToolBench
MCP server for Agent Fleet hub integration, providing tools for task management, loop governance, messaging, and agent coordination
40 tools with consistent naming (fleet_* prefix) and basic descriptions, but critical gaps in schema completeness and parameter documentation. All tools have descriptions (10-150 chars, mostly adequate), but input schemas lack depth: most parameters missing type constraints, enums, and validation rules. No output schemas documented. Error handling absent. Composition is reasonable (single responsibility per tool), but parameter descriptions often generic ('Task ID', 'Loop ID') without format/constraint guidance. No evidence of idempotency hints, permission gates, or audit logging.
Get details of a specific approval
List pending approvals for loops
Resolve an approval (approve or reject)
List all channels and their membership
Create a new channel
Create a new project
Create a new task in a project
Send a direct message to another agent (point-to-point, not broadcast)
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., loop_id, task_id) for chaining.
Parameter descriptions lack format/constraint guidance. 'Task ID to claim' does not specify format (UUID? alphanumeric?), length, or validation rules. LLMs cannot self-correct invalid input.
No error handling guidance. Tools do not document what errors are retryable, user-fixable, or fatal. No recovery hints (e.g., 'Try fleet_users() first if agent name unknown').
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | 2026-07-28+ | v2 |
Register with the hub and obtain a session token for subsequent authenticated calls
Join an existing channel
Leave a channel
Add an artifact to a loop
Add a dependency between loops
Amend the governance policy of a loop
Authorize a loop for autonomous execution
Claim a loop for execution
Create a new governed loop for autonomous task execution
Get details of a specific loop
Send a heartbeat to keep a loop alive
List loops by state
Pause a running loop
List loops ready for claiming
Request a state transition for a loop
Resolve a loop with a final verdict
Resume a paused loop
Stop a running loop
Advance a loop to the next iteration
Get the tree structure of a loop and its sub-loops
View the task board for a project
List tasks owned by the current agent
Retrieve pending messages addressed to the current agent
Send a message to a recipient or broadcast to a channel
Unregister from the hub and invalidate the session token
Claim a task for work
Mark a task as complete
Get details of a specific task
Mark a claimed task as in progress
List tasks ready for work (not yet claimed or in progress)
List all registered users and their online status
Get the current agent's identity and session information
Enum constraints missing for state/status parameters. 'Loop state filter (e.g., running, paused, completed)' should be a formal enum. LLMs may hallucinate invalid states like 'pending' or 'archived'.
No pagination or result-limit documentation. fleet_users, fleet_channels, fleet_tasks_ready accept 'limit' but do not document max, default, or whether results are truncated. Risk of context window exhaustion.