Custom Gemini with Backend API / MCP / A2A interactions

Community
1

We want to achieve below flow :

User will come to https://gemini.google.com/app .

  • User will ask user query
  • Gemini will decide if Backend Tool API / MCP Tool has to be invoked or will be served without Backend-API/ MCP tool
  • If Backend-API/ MCP tool is necessary then Backend-API/ MCP tool will be invoked and it will give response back and rended by Gemini.
  • In case of generic queries ( not served by Backend-API/ MCP tool ) , Gemini will provide response.

Note: We don’t want to implement our own chat interface. User Interaction and response rendering needs to be done in https://gemini.google.com/app or Gemini Gems only.

Basically we want our own flavored Gemini which interacts with our backend services / MCP tools.

Any help will be really appreciated.

2

@Dinesh_Vyas ,

welcome to the community,

from what you described above i think you are looking to create a chat bot/agent with MCP Tool

this cannot be achieved by a gemini app at this point.

you can use google ADK to quickly build this and test and deploy with very little effort.

you can also use BUILD feature in AI studio to speedup the development for you where it can generate the entire UI and api for you and you can deploy this with a click of a button.

1 Like
3

Yeah there’s probably a tutorial about how to do this. You could probably vibe code it in a day.

4

Hey Dinesh. That sounds really exciting. I would start by reading the offial documentation. Unlike tradition, uber-technical pieces, it is newbie and AI friendly in case you are working with an AI partner that needs to understand the documentation in detail before you start building anything.

Drop relevant urls eg. documentation on MCP, A2A, etc. to your agent’s context and ask something like “Read the latest gemini documentation about MCP, A2A, and understand how it works in ideal circumstances, what are the parameters, calls, and options we have and how to correctly use them, Do not write any code, initially give me an analysis of how everything works and what we can do with this.”

This will help you better understand conceptually what are the possibilities and if it fits your project, or if you have to adjust your project to work with how MCP, A2A actually works.

Then I would start with a basic PoC, before diving into your final app. You can start with a local Python env, node.js / vite for front-end testing, or just good old terminal view <3

The important part for using microservices / function calling / MCP tools is good context awarness. Meaning your agent(s) should be aware of the possibilities and how to correctly handle them, as well as a parsing tool or dedicated agent that understands user requests and parses for intent to use a tool or expects the results of a tool.

If you let us know more about what you are trying to achieve in terms of operations, could give more hints. Good luck with your project <3