Loading ToolBench
A private system for managing people, tasks, notes, presence, and hosting/fronting history with MCP integration for AI assistants
Bunch demonstrates solid foundational quality with 40 tools across demo and private systems. All tools have descriptions (10-200 chars, baseline 194) and explicit input schemas with type definitions. Tool naming follows verb_noun convention (get_, list_, create_, update_, delete_). However, parameter descriptions are inconsistent, many lack detail on constraints, formats, or dependencies. Output schemas are partially documented but sparse. Error handling guidance is minimal. Demo tools are well-annotated with security metadata; private tools lack equivalent rigor. Composition is sound (single responsibility per tool), but some tools could benefit from richer context (e.g., create_alter lacks guidance on when to use vs. update_alter).
Modify an existing image using ChatGPT's image editing capabilities
Sign in to use your own private Bunch system instead of the fictional Demo system. Private actions are advertised during discovery and become usable only after authorization. Existing authenticated access and tenant restrictions apply.
Create a new profile (alter) in the system
Create a new note
Create a new to-do item
Delete a profile (alter) from the system
Delete a note
Private system tools (list_alters, get_alter, create_todo, etc.) lack detailed parameter descriptions. E.g., 'alterId' is described only as 'UUID of the alter' without guidance on format, constraints, or how to obtain it. LLMs cannot infer whether to pass a raw UUID or a display name.
Output schemas are not visible in source code for private system tools. Only demo tools show explicit outputSchema definitions. Without documented return types, LLMs cannot plan downstream tool calls or extract required fields (e.g., does create_alter return the new alterId?).
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2025-06-18+ | v2 |
Delete a to-do item
Record the end of a fronting episode
Generate an image using a text prompt
Retrieve the authenticated account profile and stable account identity for the current user
Retrieve details for a specific profile (alter)
Retrieve a specific conversation summary
Get the current hosting and fronting status
Read the fictional saved-record catch-up for Benny's return. Fenton and Dot have no sample catch-up; return null for them rather than inventing one. Review state does not complete tasks.
Read one fictional person's description, profile picture, preferences, and relationships. Dot is no relation to Fenton or Benny.
Read the fixed fictional presence snapshot. Hosting is responsibility; fronting is presence and can overlap. This is not live personal presence.
Read Bunch's populated fictional Demo system: Fenton, Benny, Dot, shared tasks, separate hosting/fronting history, and a sample catch-up. Use for a demonstration or default walkthrough, never as the user's real records. Read-only; no sign-in needed.
Retrieve MCP tool usage statistics and AI spending analytics
Retrieve a specific note
Retrieve a specific to-do item
List all profiles (alters) in the private system
List saved conversation summaries and catch-up records
Explore fictional hosting/fronting periods, optionally filtered by person and kind. Open hosting continues after fronting ends. Missing records do not prove absence.
Browse fictional notes and reminders, optionally filtering by recipient/relevance or author. Fenton reminds Benny to write about the gift the system received.
Explore the fictional Demo system's people and retrieve stable demo IDs. No private records.
Browse fictional tasks, optionally filtered by Relevant to person and OPEN/DONE status. Shared relevance is separate from who handles the work and from presence.
List hosting and fronting history, optionally filtered by alter or kind
List all notes in the system, optionally filtered by alter or author
List all to-do items in the system, optionally filtered by status or alter
Prepare a catch-up session for an alter based on conversation summaries
Render an interactive native scene widget
Save a new conversation summary for catch-up
Set appearance notes and reference images for a profile
Mark a catch-up item as reviewed or completed
Set the current system host (responsible alter)
Record the start of a fronting episode for an alter
Update an existing profile (alter)
Update an existing note
Update an existing to-do item
Destructive tools (delete_alter, delete_todo, delete_note) lack confirmation or dry-run support. No error handling guidance for recovery. LLMs may delete data without understanding consequences.
Image generation tools (generate_image, alter_image) lack constraints on prompt length, image dimensions, or API rate limits. No guidance on failure modes (e.g., 'prompt too vague', 'rate limited'). LLMs cannot self-correct on invalid inputs.
Presence and fronting tools (start_fronting_episode, end_fronting_episode, set_system_host) lack documentation on state transitions and constraints. E.g., can you end an episode that hasn't started? Can multiple alters front simultaneously? Undocumented state rules cause silent misuse.