Loading ToolBench
Client-side, backend-free AI assistant platform for OpenAI and xAI Grok with tool calling, MCP servers, web/X search, image generation, TTS, and fully local storage.
Wordmark is a client-side web application, not an MCP server. It integrates WITH MCP servers and external APIs (OpenAI, xAI, LM Studio, Ollama) but does not itself expose an MCP interface. The 11 'tools' listed are either pass-through wrappers to external APIs (open_meteo_forecast, web_search, code_interpreter, image_generation, shell, file_search, grok_generate_image, grok_edit_image, openai_generate_image, openai_edit_image) or internal agent state management (queue_followup). Most tools lack visible input schemas in the provided source; only 2 tools (open_meteo_forecast, grok_generate_image, grok_edit_image, openai_generate_image, openai_edit_image) show explicit schemas. Descriptions are present but generic. No error handling guidance, no output schemas documented, no parameter validation rules visible. This is a consumer application, not a production MCP server.
Allow the assistant to run Python code and work with files in the provider sandbox.
Search through uploaded documents using vector stores.
Edit one or more images with xAI Grok Imagine. If no image URL is provided, the most recent uploaded or generated image is used.
Generate an image with xAI Grok Imagine.
Generate or edit images with OpenAI. Uses the built-in image tool on OpenAI and the gpt-image-2 API (OpenAI API key required) on other services.
Get a short weather forecast via Open-Meteo (1-7 days).
Not an MCP server. Wordmark is a client-side web application (Vite + Electron) that consumes MCP servers and external APIs. It does not expose an MCP interface itself. Evaluation against MCP server rubric is categorically invalid.
7 of 11 tools lack visible input schemas in source code (web_search, code_interpreter, image_generation, shell, file_search).
Descriptions for 7 tools are generic placeholders (35-40 chars) lacking actionable context. 'Allow the assistant to...' does not explain WHEN to use the tool, what it returns, or prerequisites. Baseline: 194 chars average for A+ tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Edit one or more images with OpenAI gpt-image-2. If no image URL is provided, the most recent uploaded or generated image is used.
Generate an image with OpenAI gpt-image-2.
Queue follow-up instructions for yourself. Each step is sent back as its own turn, in order, once the current turn ends. Write each one as a self-contained instruction — the turn that receives it sees the conversation, not your reasoning for scheduling it. Queue only work you are NOT doing in this turn: a step you go on to carry out here comes back to you anyway, and you will have been handed work that is already finished. The queue survives the turn, so call this once with what is genuinely left rather than re-listing the plan every turn; steps repeating something already queued or already run are refused, and the result tells you which. Do not use it to ask the user questions.
Allow the assistant to run shell commands in a sandboxed container environment.
Allow the assistant to use provider-managed web searches for fresh information on OpenAI or xAI.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure. Baseline: 100% of A+ tools document return types.
No error handling guidance visible. Tools that call external APIs (weather, image generation, shell) can fail in many ways, but no recovery hints provided. LLMs receive raw errors with no actionable next steps.