NanoBananaPro

GenAI MCP Server

This project implements a Model Context Protocol (MCP) server for image generation and image editing using Google Gemini (via google.golang.org/genai) and Tongyi Wanxiang (Ali Bailian) image APIs, plus optional automatic upload of generated images to S3‑compatible object storage (AWS S3, Aliyun OSS, etc.).

The server exposes a streamable HTTP MCP endpoint and provides tools for Gemini and Wan:

• gemini_generate_image – text → image

• gemini_edit_image – image + text → edited image

Gemini / Nano Banana backend support

This MCP server currently supports the following Gemini‑compatible backends:

1. Google official Gemini API

   • Use the default GENAI_BASE_URL=https://generativelanguage.googleapis.com

   • GENAI_API_KEY is a Google Gemini API key

2. dmxapi (Gemini‑compatible third‑party gateway)

   • Set GENAI_BASE_URL to the dmxapi Gemini endpoint (for example https://www.dmxapi.cn)

   • GENAI_API_KEY is the key issued by dmxapi

   • As long as the endpoint implements the google.golang.org/genai compatible Gemini API, no code changes are needed

Tongyi Wanxiang (Ali Bailian) backend support

When GENAI_PROVIDER=wan, the server will use Ali Bailian Tongyi Wanxiang image APIs (via DashScope) instead of Gemini:

• Set:

  • GENAI_PROVIDER=wan

  • GENAI_BASE_URL=https://dashscope.aliyuncs.com

  • GENAI_API_KEY=<your DashScope API key>

  • GENAI_GEN_MODEL_NAME=wan2.5-t2i-preview (text → image)

  • GENAI_EDIT_MODEL_NAME=wan2.5-i2i-preview (image → image)

• Wan provides a separate MCP tool set (see internal/tools/wan.go):

  • wan_create_generate_image_task

  • wan_query_generate_image_task

  • wan_create_edit_image_task

  • wan_query_edit_image_task

The Python test client in tests/mcp_client.py will automatically route calls to Gemini or Wan based on GENAI_PROVIDER (gemini by default, wan for Tongyi Wanxiang).

1. Prerequisites

• Go 1.21+ (recommended; go.mod uses module mode)

• A valid Gemini API key

• Optional: S3 / OSS bucket for storing images

2. Configuration (.env)

Copy env.example to .env, then fill in real values.

GenAI configuration

HTTP server

MCP endpoint will listen on:

OSS / S3 configuration (optional, required when

When GENAI_IMAGE_FORMAT=url:

• For Aliyun OSS: make sure

  • OSS_ENDPOINT is like oss-cn-beijing.aliyuncs.com

  • The bucket policy allows read access if you expect the returned URL to be publicly accessible

3. Running the MCP Server

You can run the MCP server in two ways:

1. Clone & build from source

   • Clone this repo and enter the project root

   • Copy env.example to .env and fill in your configuration

   • Run:

     go build . ./genai-mcp

2. Download release binary

   • Download the appropriate binary from the Releases page

   • Place it in a directory of your choice

   • Copy env.example from this repo (or from the release asset) to .env in the same directory and update configuration

   • Run (binary name may vary by platform):

     ./genai-mcp

By default the MCP HTTP endpoint will be:

You can connect to this MCP endpoint from any MCP‑compatible client (e.g. Code editors or tools that support the streamable HTTP MCP transport).

4. MCP Tools

The server registers two tools in internal/tools/gemini.go:

• gemini_generate_image

  • Input:

    • prompt (string, required): text prompt describing the image

  • Output:

    • When GENAI_IMAGE_FORMAT=base64: a data:image/...;base64,... string

    • When GENAI_IMAGE_FORMAT=url: an OSS/S3 URL generated by the server

• gemini_edit_image

  • Input:

    • prompt (string, required): how to edit the image

    • image_url (string, required): original image URL or data URI

  • Output:

    • Same format as above (base64 or url), depending on configuration

When GENAI_IMAGE_FORMAT=url:

• Generated / edited images are:

  • Downloaded (if Gemini returns a URL), or decoded (if it returns inline data)

  • Re‑uploaded to OSS / S3

  • Stored under key pattern: images/yyyy-MM-dd/{uuid_timestamp_random}.ext

5. Contact

• WeChat: Scan the QR code below to add as a friend

  

• Discord: Username adamydwang

Star History