Universal AI-to-Tool bridge: MCP+A2A dual protocol gateway for CLI, Office, and Browser adapters
AI-Bridge exhibits significant definition quality gaps across 23 tools. While tool names follow verb_noun conventions (aibridge_interact, browser_navigate, gimp_open_image), descriptions vary wildly in quality and language. Many descriptions are in Chinese, creating accessibility issues. Critical gaps: (1) aibridge_interact lacks parameter descriptions for 'target' object properties (name, role, xpath, css, automation_id); (2) browser_* tools have Chinese descriptions without English equivalents; (3) no output schemas documented for any tool; (4) error handling guidance absent; (5) many tools lack actionable parameter constraints. The 'target' parameter in aibridge_interact is a complex nested object with no type information visible for its properties. Tools like browser_execute_intent and browser_execute_task accept free-form natural language without validation constraints. No tool declares permissions, audit trails, or security boundaries.
Get the status of a specific application
Check health status of all adapters
Interact with GUI applications. Available apps: chrome, edge, word, excel, powerpoint, wps_writer, wps_spreadsheet
List all available applications and their capabilities
Add a 3D object to the scene
Render the scene to an image
Create a new Blender scene
点击页面上的元素,支持多种定位方式
No output schemas documented for any of 23 tools. LLMs cannot predict response structure, forcing them to guess at field names and types for downstream tool chaining.
Browser tools (browser_navigate, browser_click, browser_type, browser_extract, browser_screenshot) have descriptions in Chinese only. English descriptions are required for international LLM compatibility and accessibility.
aibridge_interact's 'target' parameter is a complex nested object with properties (name, role, xpath, css, automation_id) that lack type definitions and descriptions. LLMs cannot determine which property to use or how to structure the object.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
使用自然语言描述意图,AI-Bridge 自动执行相应操作。例如:'搜索iPhone 15'、'点击提交按钮'、'提取所有商品名称'
执行复杂的多步骤任务,使用O-R-A(观察-推理-行动)循环自动完成。例如:'在京东上搜索iPhone 15,提取前3个商品的价格和评分'
从页面提取结构化数据
导航浏览器到指定URL,并返回页面基本信息
截取当前页面或指定元素的截图
在输入框中输入文本
Convert video/audio format
Extract audio from video
Trim/cut video segment
Apply a filter effect to the current image
Export the current image to a file
Open an image file in GIMP
Export the project to a video file
Create a new video editing project
Add a video/audio clip to the timeline
browser_execute_intent and browser_execute_task accept free-form natural language ('intent', 'goal') without validation constraints, enums, or length limits. No guidance on what intents are supported or how complex tasks should be structured.
No error handling guidance across any tool. Tools do not document retryable vs fatal errors, recovery steps, or what invalid inputs should return. LLMs have no recovery path on failure.