MCP server for MIMO image recognition
Wellknown found it in public sources; nobody has proven control of it yet. Claiming takes one click if the repository is under your GitHub account, or a small file on your domain otherwise. Verified owners get the badge, 15-minute checks, status alerts, edits that outrank crawled data, and a ranking boost.
Agents can do it too: POST https://wellknown.network/api/v1/claims with {"agent":"mimo-image-recognition-mcp","method":"well_known_file"} — machine-readable steps at claim.json, guide at /docs/claim.
Everything here was measured by our prober or read from a registry. Nothing is self-reported.
Attributed to the source that supplied each field. Treated as claims, not facts.
# MIMO Image Recognition MCP 一个调用小米 MIMO 多模态模型进行图片理解的 MCP Server。 该 MCP 支持: - 本地图片路径识别 - 网络图片 URL 识别 - Agent 自定义提示词 - API Key、请求地址、模型名称通过 MCP 启动配置传入 - 支持通过 PyPI / uvx 运行 - 支持本地源码运行 ## 功能说明 本项目会向 MCP 客户端暴露一个工具: ### `understand_image` 用于调用 MIMO 多模态模型理解图片。 支持的输入方式: - `image_path`: 单张本地图片路径 - `image_url`: 单张网络图片 URL - `image_paths`: 多张本地图片路径 - `image_urls`: 多张网络图片 URL - `prompt`:由 Agent 自己决定的图片理解任务 - `system_prompt`:可选系统提示词 - `temperature`:输出随机性 - `max_tokens`:最大输出长度 --- ## 安装方式一:通过 PyPI / uvx 使用 如果你只是想使用这个 MCP,推荐使用这种方式。 MCP 配置示例: ```json { "mcpServers": { "mimo-image-recognition": { "command": "uvx", "args": [ "--refresh", "mimo-image-recognition-mcp" ], "env": { "MIMO_API_KEY": "用户自己的 API Key", "MIMO_API_BASE": "https://token-plan-cn.xiaomimimo.com/v1", "MIMO_MODEL": "mimo-v2.5" } } } } ``` 配置项说明: | 配置项 | 说明 | |---|---| | `MIMO_API_KEY` | 你的 MIMO API Key | | `MIMO_API_BASE` | MIMO API 请求地址,通常为 `https://api.xiaomimimo.com/v1`或`https://token-plan-cn.xiaomimimo.com/v1` | | `MIMO_MODEL` | 要调用的 MIMO 模型名称,例如 `mimo-v2.5` | --- ## 安装方式二:本地源码运行 如果你想修改源码或参与开发,可以使用本地源码方式。 ### 1. 克隆项目 ```bash git clone https://github.com/Mriestac/mimo-image-recognition-mcp.git cd mimo-image-recognition-mcp ``` ### 2. 安装依赖 ```bash uv sync ``` ### 3. MCP 配置示例 ```json { "mcpServers": { "mimo-image-recognition": { "command": "uv", "args": [ "--directory", "<你的项目路径>", "run", "mimo-image-recognition-mcp" ], "env": { "MIMO_API_KEY": "你的 MIMO API Key", "MIMO_API_BASE": "https://token-plan-cn.xiaomimimo.com/v1", "MIMO_MODEL": "mimo-v2.5" } } } } ``` 请把 `<你的项目路径>` 改成你自己本地项目的真实路径。 --- ## 本地调试 可以使用 MCP Inspector 调试: ```bash uv run mcp dev src/mimo_image_recognition_mcp/server.py ``` 如果能看到: ```text understand_image ``` 说明 MCP Server 启动成功。 如果你想在 MCP Inspector 中实际调用 MIMO 接口,可以在当前终端临时设置: ### Windows …
Mapped onto the structured taxonomy from declared text and observed tool names. Confidence shown for derived entries.
Every source is kept verbatim. Field changes are logged as events.