Loading ToolBench
Unified MCP server for Vutler workspace and agent operations, providing tools for email, file management, calendar, memory, shell execution, webhooks, and Nexus node skill execution
Vutler exposes 22 tools with mixed quality. Most tools (1-7, 17-22) lack visible input schemas in the provided source, only tools 8-16 show explicit JSON Schema definitions. Descriptions vary: some are detailed (send_email, draft_email, read_emails), others are generic (web_search, email, drive, calendar, memory, shell, webhook). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present. Error handling and recovery guidance are absent. Parameter descriptions exist for tools 8-16 but are missing for tools 1-7. Output schemas are not documented. The server mixes high-risk tools (shell, open_terminal_session) with low-risk ones without clear permission boundaries.
Manage calendar events and schedules
Close an existing Nexus terminal session when it is no longer needed.
Create an email draft for later review or manual sending. Use source='google' or source='microsoft365' only when the user explicitly wants a personal Gmail / Outlook mailbox; that path is treated as on-behalf mail and stays approval-gated.
Access and manage files in VDrive
Send and receive emails
Send input to an existing Nexus terminal session and return incremental output.
Tools 1-7 (web_search, email, drive, calendar, memory, shell, webhook) lack visible input schemas in source code.
Generic, under-20-character descriptions for tools 1-7 (e.g., 'Send and receive emails', 'Access and manage files').
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) present. High-risk tools (shell, open_terminal_session, exec_terminal_session) are not marked destructive, preventing LLM safety reasoning.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Execute a specific agent skill on the local Nexus node
List files and folders in a directory on the user's computer.
Store and retrieve agent memories
Open a persistent terminal session on the Nexus node in a specific working directory.
Read upcoming calendar events from the user's calendar app.
Read the current clipboard content from the user's computer.
Search the user's contacts (Apple Contacts or Outlook).
Read the content of a document on the user's computer (PDF, DOCX, XLSX, CSV).
Read recent emails from the selected mailbox source. Use source='local' for Nexus Local desktop mail, source='google' for Gmail, source='microsoft365' for Outlook / Microsoft 365, or source='workspace' for the Vutler workspace mailbox. If the user says only 'check email' and multiple sources are available, ask which source they mean before calling this tool.
Read new output from an existing Nexus terminal session using the last known cursor.
Search for files on the user's local computer by name or content.
Send email using the current agent's provisioned mailbox by default. Use source='google' or source='microsoft365' only when the user explicitly says to write on their behalf or from their own Gmail / Outlook; that path always requires approval. If the user says only 'on my behalf' and multiple personal mailbox sources are available, ask which one before calling this tool.
Execute shell commands (admin only)
Inspect the current state of a Nexus terminal session, including cwd and closed status.
Search the web for information
Call external webhooks
No error handling or recovery guidance documented. Tools lack actionable error messages (e.g., 'User not found. Try search_users() first'). LLM cannot self-correct on failures.
Output schemas not documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does send_email return? success boolean? message_id?).