Loading ToolBench
A Node.js backend server for the WEAM platform that provides MCP tools for image generation (DALL-E, Gemini), web search, and integrations with various AI models and external services. Supports LangChain tool integration and multiple LLM providers.
WEAM exposes 3 tools with significant quality gaps. All three tools have descriptions, but they are verbose (300+ chars) and contain implementation details (S3 uploads, API keys) that belong in docs, not LLM-facing descriptions. Schemas are present but lack proper parameter descriptions, 'query', 'prompt', 'text' are aliases with no explanation of when to use each. No output schema is documented. Error handling is absent. The gemini_image_generator and dalle_api_wrapper descriptions include instruction text ('DO NOT use this tool if...', 'Always include...') that should be enforced via tool composition, not embedded in descriptions. searxng_search has a minimal description ('Search the web with SearxNG') that fails to explain when to use it vs. other search tools or what structure it returns. No tool declares permissions, audit requirements, or error recovery paths.
An image generation tool that creates high-quality images from text descriptions using OpenAI's DALL-E 3 model. The tool automatically uploads generated images to S3 storage and returns S3 URLs for immediate display. Tool supports various image sizes and aspect ratios, including 1024x1024 for Square images, 1024x1536 for Portrait images and 1536x1024 for Landscape images. 1024x1024 (Square): Ideal for social media posts, profile pictures, digital artwork, and product images. 1024x1536 (Portrait): Perfect for mobile content, social media stories, and vertical ads. 1536x1024 (Landscape): Great for presentations, video thumbnails, website banners, and widescreen displays. IMPORTANT: This tool automatically handles S3 uploads and returns S3 URLs for better user experience. DO NOT use this tool if the user requests to generate code based on an image input and a prompt. For such cases, use the chat tool to generate code from the image and prompt.After using this tool, always include the tool result (e.g. image URLs or Markdown) in your final response in this format .
Image generation tool powered by Google's Gemini 2.5 Flash Image model. This tool creates high-quality, photorealistic images from text descriptions using Gemini's advanced multimodal capabilities. The tool automatically uploads generated images to S3 storage and returns S3 URLs for immediate display. Usage Patterns: - Image Generation: "Create a photorealistic portrait of...", "Generate an image showing..." - Style Operations: "Create in watercolor style...", "Make it look futuristic..." - Scene Creation: "Show a landscape with...", "Create an interior scene..." IMPORTANT: Always call this tool for image generation query, do not consider the message history. This tool automatically handles S3 uploads and returns S3 URLs for better user experience. Always use the S3 url returned from this tool in your output for displaying the generated image. DO NOT use this tool if the user requests to generate code based on an image input and a prompt. For such cases, use the chat tool to generate code from the image and prompt.After using this tool, always include the tool result (e.g. image URLs or Markdown) in your final response in this format .
Descriptions embed implementation details and instruction text instead of LLM-optimized summaries. gemini_image_generator and dalle_api_wrapper descriptions include 'DO NOT use this tool if...', 'Always include...', and S3 upload mechanics. These belong in docs or tool composition, not in the description field that LLMs use for selection.
Parameter descriptions are missing or trivial. 'query', 'prompt', 'text' are listed as aliases with no explanation of which to prefer or when each applies. 'apiKey' is marked optional but no guidance on fallback behavior. searxng_search 'query' has only 'Search query string', no format, length, or example constraints.
No output schemas documented. Callers cannot know what fields to expect from image generation (URL format? metadata? error details?) or search results (rank? snippet? source?). This forces LLMs to guess and wastes tokens on parsing unstructured responses.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 33 | <=2025-11-25 | v2 |
Search the web with SearxNG
No error handling or recovery guidance. If image generation fails, API key is invalid, or search returns no results, the LLM has no actionable error message or next step. Descriptions do not mention prerequisites, rate limits, or failure modes.
API keys exposed as optional tool parameters. 'apiKey' parameter in image tools violates secret-injection pattern, credentials should never appear in tool params, as they are logged in traces and prompt history. Use server-side environment variable injection instead.