Skip to main content
Glama
Sign In
API易 MCP Server

API易 MCP Server

by 2799662352
OverviewSchemaSourceRelated ServersDiscussionsAdmin
JavaScript
Remote

API易 MCP Server

Docker Hub GitHub

基于 aistudio-mcp-server 改造，支持 API易 的 Gemini 原生格式调用。

Docker Hub: zuozuoliang999/apiyi-mcp-server

✨ 特性

  • 🎬 视频理解：支持 MP4, AVI, MOV, WebM, FLV, MPG, WMV (最大 20MB)

  • 🎙️ 音频处理：支持 MP3, WAV, AIFF, AAC, OGG, FLAC

  • 🖼️ 图片分析：支持 JPG, PNG, GIF, WebP, SVG, BMP, TIFF

  • 📄 文档转换：PDF 转 Markdown，支持 TXT, MD, JSON, XML, CSV, HTML

  • 🔍 Google 搜索：内置 Google 搜索能力

  • 💻 代码执行：支持 Python 代码执行

  • 🧠 思维链推理：支持 Gemini 2.5 系列的 thinking_budget

  • 💰 分辨率优化：支持媒体分辨率调整，节省 tokens 费用

  • 🐳 Docker 支持：完整的 Docker 容器化部署

Related MCP server: AI Vision MCP Server

🚀 快速开始

1. 获取 API易 密钥

  1. 访问 API易控制台 注册账号

  2. 充值后获取 API Key

2. 配置 MCP 客户端

Cursor / Claude Desktop 配置

{
  "mcpServers": {
    "apiyi": {
      "command": "node",
      "args": ["D:/jianji_FFMPEG/apiyi-mcp-server/dist/index.js"],
      "env": {
        "APIYI_API_KEY": "你的API易密钥",
        "GEMINI_MODEL": "gemini-3.1-pro-preview-thinking",
        "APIYI_BASE_URL": "https://api.apiyi.com"
      }
    }
  }
}

使用 npx 方式（需要先发布到 npm）

{
  "mcpServers": {
    "apiyi": {
      "command": "npx",
      "args": ["-y", "apiyi-mcp-server"],
      "env": {
        "APIYI_API_KEY": "你的API易密钥"
      }
    }
  }
}

3. 🐳 Docker 部署 (推荐)

无需安装 Node.js，直接从 Docker Hub 拉取镜像！

# 直接使用 Docker Hub 镜像
docker pull zuozuoliang999/apiyi-mcp-server:latest

# 运行容器
docker run -it --rm \
  -e APIYI_API_KEY=你的API易密钥 \
  -v /path/to/your/media:/app/media:ro \
  zuozuoliang999/apiyi-mcp-server:latest

使用 docker-compose

# 克隆项目
git clone https://github.com/2799662352/apiyi-mcp-server.git
cd apiyi-mcp-server

# 创建环境变量文件
echo "APIYI_API_KEY=你的API易密钥" > .env

# 启动服务
docker-compose up -d

# 查看日志
docker logs -f apiyi-mcp-server

Cursor MCP 配置 (Docker 方式)

{
  "mcpServers": {
    "apiyi": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "APIYI_API_KEY=你的API易密钥",
        "-v", "/path/to/your/media:/app/media:ro",
        "zuozuoliang999/apiyi-mcp-server:latest"
      ]
    }
  }
}

Windows + Cursor 完整实战配置（含长输出 / 长超时 / Gemini 3.1 思维链）

下面这份是经过实测可用的 Windows 配置：把整个 D:\ 盘以只读方式挂进容器内的 /app/media，这样 Cursor 里给 files.path 传 D:/video/foo.mp4 也能直接被容器读到；同时通过 -e 单独声明环境变量（key 从 env 字段注入，避免在 args 里出现明文密钥）。

{
  "mcpServers": {
    "apiyi": {
      "command": "docker",
      "args": [
        "run", "-i", "--rm",
        "-e", "APIYI_API_KEY",
        "-e", "GEMINI_MODEL",
        "-e", "APIYI_BASE_URL",
        "-e", "GEMINI_MAX_OUTPUT_TOKENS",
        "-e", "GEMINI_TIMEOUT",
        "-v", "D:/:/app/media:ro",
        "zuozuoliang999/apiyi-mcp-server:latest"
      ],
      "env": {
        "APIYI_API_KEY": "sk-YOUR_API_KEY_HERE",
        "GEMINI_MODEL": "gemini-3.1-pro-preview-thinking",
        "APIYI_BASE_URL": "https://api.apiyi.com",
        "GEMINI_MAX_OUTPUT_TOKENS": "65536",
        "GEMINI_TIMEOUT": "1800000"
      }
    }
  }
}

要点说明：

  • -v D:/:/app/media:ro：把整个 D: 盘只读挂载到容器，Cursor 中无论传 D:/... 还是 D:\... 路径都能命中容器内 /app/media/...。

  • 用 -e KEY + env.KEY 的分离写法，可避免把密钥明文写在 args 中（args 在某些日志里会被打印）。

  • GEMINI_MAX_OUTPUT_TOKENS=65536：跑长视频/长文档分析时建议放开，默认 8192 容易被截断。

  • GEMINI_TIMEOUT=1800000（30 分钟）：搭配 thinking 模型/大文件多模态分析时避免提前超时。

  • GEMINI_MODEL=gemini-3.1-pro-preview-thinking：开启 Gemini 3.1 思维链模型；如想节省成本可改回 gemini-2.5-flash。

4. 本地开发 (需要 Node.js)

# 进入项目目录
cd D:\jianji_FFMPEG\apiyi-mcp-server

# 安装依赖
npm install

# 编译
npm run build

# 启动
npm start

⚙️ 环境变量配置

变量名

描述

默认值

APIYI_API_KEY

API易 密钥 (必需)

-

GEMINI_API_KEY

备用：Google AI Studio 密钥

-

APIYI_BASE_URL

API易 端点

https://api.apiyi.com

GEMINI_BASE_URL

备用：自定义端点

-

GEMINI_MODEL

默认模型

gemini-3.1-pro-preview-thinking

GEMINI_TIMEOUT

超时时间 (ms)

300000 (5分钟)

GEMINI_MAX_OUTPUT_TOKENS

最大输出 tokens

8192

GEMINI_MAX_FILES

最大文件数

10

GEMINI_MAX_TOTAL_FILE_SIZE

最大总文件大小 (MB)

50

GEMINI_TEMPERATURE

温度参数

0.2

GEMINI_MEDIA_RESOLUTION

媒体分辨率 (LOW/MEDIUM/HIGH)

MEDIUM

📖 使用示例

视频理解

{
  "user_prompt": "分析这个视频的内容，描述主要场景和人物动作",
  "files": [
    {
      "path": "D:/jianji_FFMPEG/误解向剪辑.mp4"
    }
  ]
}

图片分析

{
  "user_prompt": "描述这张图片中的内容",
  "files": [
    {
      "path": "/path/to/image.jpg"
    }
  ]
}

PDF 转 Markdown

{
  "user_prompt": "将这个 PDF 转换为格式良好的 Markdown，保留结构和格式",
  "files": [
    {"path": "/document.pdf"}
  ]
}

Google 搜索

{
  "user_prompt": "2024年最新的AI技术突破有哪些？",
  "enable_google_search": true
}

代码执行

{
  "user_prompt": "编写并运行一个计算100以内质数的Python脚本",
  "enable_code_execution": true
}

思维链推理 (Gemini 3.1 Thinking)

{
  "user_prompt": "详细分析量子计算的技术方案",
  "model": "gemini-3.1-pro-preview-thinking"
}

Gemini 3.1 思维链版本默认开启 thinking，无需手动指定 thinking_budget。若你仍想精细控制，可继续传 thinking_budget 参数。

💰 媒体分辨率优化 (节省费用)

使用较低分辨率处理图片/视频，节省 tokens 费用：

{
  "user_prompt": "这张图片的主题是什么？",
  "files": [{"path": "/path/to/large_image.jpg"}],
  "media_resolution": "LOW"
}

分辨率

说明

适用场景

LOW

最低分辨率

简单识别、节省费用

MEDIUM

中等分辨率 (默认)

一般场景

HIGH

最高分辨率

需要细节分析

📚 API易 文档参考

🔧 支持的模型

模型

描述

gemini-3.1-pro-preview-thinking

默认：Gemini 3.1 思维链版本，最强复杂任务/长上下文

gemini-3.1-pro-preview

Gemini 3.1 标准版（不开 thinking，速度更快）

gemini-3-pro-preview

Gemini 3 预览版（上一代）

gemini-2.5-flash

混合推理，响应最快，适合简单识别

⚠️ 注意事项

  1. 文件大小限制：单个媒体文件不超过 20MB

  2. 推理成本：thinking tokens 会计入输出成本

  3. 代码执行：仅支持 Python，在沙箱环境中运行

  4. API 密钥：使用 API易 密钥，非 Google AI Studio 密钥

🔗 链接

📄 许可证

MIT License

基于 eternnoir/aistudio-mcp-server 修改

Available Tools

2 tools
generate_contentA

Generate content using Gemini with optional file inputs, code execution, and Google search. Supports multiple files: images (JPG, PNG, GIF, WebP, SVG, BMP, TIFF), video (MP4, AVI, MOV, WebM, FLV, MPG, WMV), audio (MP3, WAV, AIFF, AAC, OGG, FLAC), documents (PDF), and text files (TXT, MD, JSON, XML, CSV, HTML). MIME type is auto-detected from file extension.

Example usage:

{
  "user_prompt": "Analyze this video",
  "files": [{"path": "/path/to/video.mp4"}]
}

PDF to Markdown conversion:

{
  "user_prompt": "Convert this PDF to well-formatted Markdown",
  "files": [{"path": "/document.pdf"}]
}

With Google Search:

{
  "user_prompt": "What are the latest AI breakthroughs in 2024?",
  "enable_google_search": true
}

With Code Execution:

{
  "user_prompt": "Write and run a Python script to calculate prime numbers",
  "enable_code_execution": true
}

Media Resolution Optimization (save tokens):

{
  "user_prompt": "Describe this image",
  "files": [{"path": "/image.jpg"}],
  "media_resolution": "LOW"
}

Inputs: user_prompt (required), files, model, temperature, +5 more.

generate_content_batchA

Generate multiple contents concurrently using Gemini. This tool allows you to send multiple requests in parallel for better performance.

Each request in the batch requires a unique 'id' field to identify results. All other parameters are the same as generate_content.

Example usage:

{
  "requests": [
    {
      "id": "req1",
      "user_prompt": "Describe image 1",
      "files": [{"path": "/image1.jpg"}]
    },
    {
      "id": "req2", 
      "user_prompt": "Describe image 2",
      "files": [{"path": "/image2.jpg"}]
    },
    {
      "id": "req3",
      "user_prompt": "Summarize this document",
      "files": [{"path": "/doc.pdf"}]
    }
  ],
  "max_concurrency": 5
}

The response includes success/failure status for each request:

{
  "total": 3,
  "succeeded": 2,
  "failed": 1,
  "results": [
    {"id": "req1", "success": true, "content": "..."},
    {"id": "req2", "success": true, "content": "..."},
    {"id": "req3", "success": false, "error": "..."}
  ]
}

Inputs: requests (required), max_concurrency.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Deploy Server
    One endpoint for every MCP toolRayrun runs every MCP server behind one URL and holds the credentials. Each call is decided per client, tool, and argument, then recorded.ray.runAd

    TDQS

    A4.1/5.0

    Scored across 2 tools

    Disambiguation5/5

    The two tools have clearly distinct purposes: one for single content generation, one for batch processing. No overlap or ambiguity.

    Naming Consistency5/5

    Both tools follow the exact same prefix 'generate_content', with a descriptive suffix '_batch' for the batch variant. Perfectly consistent pattern.

    Tool Count4/5

    With only two tools, the server is minimal but scoped appropriately for Gemini content generation. Could potentially include a model listing tool, but not necessary.

    Completeness4/5

    The tool set covers the core functionality (single and batch generation). Missing optional features like model selection or streaming, but not gaps that prevent basic usage.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Resources

    Related MCP Connectors

    Related MCP Servers