Loading ToolBench
MCP endpoint for Claude, agent tunnel, telemetry store, and operator dashboard
kenny-server has 52 tools with mostly complete schemas and descriptions, but significant gaps in parameter documentation and error handling guidance. Tool naming is generally clear (verb_noun pattern), but many parameters lack type constraints and descriptions. Output schemas are not documented. Error recovery guidance is minimal. The server handles high-risk operations (account creation, shell execution, webfilter configuration) but lacks confirmation patterns and detailed permission documentation.
Create a new user account.
Delete a user account.
Perform an action on a user session.
Set admin privileges for a user account.
Enable or disable a user account.
Set logon rights for a user account.
Per-section health status/summary for one agent.
Latest stored telemetry snapshot for an agent (optionally one section).
Missing input schemas for 8 tools (ticket_draft, ticket_find, webfilter_set, web_activity_query, reliability_suppression_add/remove, ticket_rule_set/remove). Cannot verify parameter types or constraints.
No output schemas documented for any tool. LLMs cannot predict response structure, forcing them to parse unstructured output and plan downstream calls blindly.
Destructive operations (account_create, account_delete, shell_exec, powershell_exec, winget_install, agent_update) lack confirmation/dry-run patterns. Agents can irreversibly modify systems without safeguards.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | B | 73 | <=2025-11-25 | v2 |
Update the agent software.
List autostart programs.
Query event log entries.
List running processes.
List system services.
Per-agent rolled-up health for the whole fleet.
Get disk usage information.
List files and directories at a path.
Read file contents.
Search for files matching a pattern.
List known agents with online state and rolled-up health.
Reset a network adapter.
Get network configuration.
Flush DNS cache.
Set password policy.
Execute PowerShell script on Windows agent.
Add a reliability alarm suppression.
List reliability alarm suppressions.
Remove a reliability alarm suppression.
Start a remote help session.
Get remote help session status.
Stop a remote help session.
Capture the agent's screen.
Set the active agent that capability tools forward to. Call this before any capability tool.
Execute shell command on Linux/macOS agent.
Collect telemetry from the agent.
Propose a ticket for the operator to correct and submit through the ordinary create route.
Look for a ticket that already exists.
List auto-ticket rules.
Remove an auto-ticket rule.
Set an auto-ticket rule.
Record on the ticket what this turn found or changed, in one or two sentences. Call this once, at the end of a turn where you learned something about the machine, changed something on it, or reached a conclusion -- and not at all for a turn that was only small talk or a question back to the person. This is what somebody reading the ticket later sees; the conversation itself stays where it was had.
Record the verdict of this triage investigation and finish. Call this exactly once, as the last thing you do. The server decides what happens to the ticket; your job is to state what you found and what proves it.
Query web browsing activity.
Apply web filter rules.
Clear web filter rules.
Get current web filter configuration.
Push configured web filter list to the host.
Set web filter configuration.
Get web filter status.
Install a package using winget.
List installed packages on Windows.
Uninstall a package using winget.
Update an installed package.
No error recovery guidance in descriptions. Tools do not explain what to do on failure (retry, ask user, abort). LLMs cannot self-correct or escalate intelligently.
No permission/scope declarations on any tool. Cannot audit which capabilities require what privileges or configure least-privilege agent access.