Using open source models from Hugging Face in Firebase studio

How can I get the firebase studio prototyper agent to be able to ground with google search so that it can read details about online documentation regarding open source models and integrating them into what I am working on? Was wondering if someone could give me some insights.

I think you might be able to use an MCP server for this, but you might have to experiment.

I have only used the firebase MCP server. I assume hugging face has an MCP server you might be able to leverage, iirc Github does, so it seems like a fairly standard thing.

1 Like

Thanks let me check it out!

Does adding the firebase MCP to the prototyper agent make it better? Could you give me a short explanation of what exactly that does? Sorry I am a firebase studio noob.

In my option it won’t help much, the prototyper is a very specific mode, more like a “vibe coding” thing. While i am not sure what the intention of it is my experience suggests that its mostly for testing things out and then when you want to do more serious production you will move your app over to the regular development tools, which are not Agentic by default.

Also the prototyper builds very “bloated” apps, it includes a lot of stuff that make it easier to produce good looking results but use a lot of resources, such as react and next.

MCP ( Model Context Protocol ) is an industry standard that is becoming quite popular, it allows you to expose an API to the AI Agent that it can use as a set of tools that preform various functions.

This can address a lot of the tool chain integration issues that come up with the current AIs.

There are tons of MCP servers out there now that can do almost anything.

For example, in the case of the Firebase MCP server, the AI can look at storage or the database in the backend, do reads and writes etc.

In the case of Githubs MCP server for MCP servers it will let you and the AI browse and configure MCP servers on Github for use in your project.

So, what it does is give the AI access to more resources but it will not improve the AI prototyper that much, and could be a bit dangerous in Agentic mode.

1 Like

Thankyou thats very helpful to know. You mentioned that the prototyper builds very bloated apps. Is there anyway to reduce this bloat? TBH I don’t have much coding experience but I can tell that there is a lot more code then there needs to be. Idk if there are AI tools or things I can do within firebase studio to reduce the bloat.

ALso I notice the prototyper dosnt always follow safety protocols with API keys and secrets. IS there a way to scan my entire repository to make sure Im not leaking my secrets?

Again thanks for the help.This is very insightful for a beginner like me.

This is just my experience and it might not be the best way to do things, what i did was use the prototyper when starting with firebase studio, then once i got used to things and had a better understanding of what i wanted i created a github template repo, using a minimal vite template, that in my case was basically a webgl canvas and some ui panels for testing database storage access.

With this i am able to create workspaces by importing the github repo as my template, which is an option when you create a workspace.

This gives you the normal AI which is not Agentic and tends to be easier to work with.

This lets me build on top of that for various applications.

As far as secrets, i agree, the protoypers is very sloppy with them, i learned to create a .env where i put all my API keys then in a file i normally call firebase.js i import them as env variables.

Many times i have had to stop the AI from thinking i don’t have API keys and trying to get me to put them in directly in the code. I think its some kind of trained habit it has.

But as far as reducing the bloat, my only solution to that was to create new leaner apps.

The prototyper seems to “need” all those tools.

1 Like

Again thanks a lot thats really helpful!

image
{

"mcpServers": {

  "firebase": {

    "command": "npx",

    "args": \["-y", "firebase-tools@latest", "mcp"\]

  }

}

}

That was the config I used provided by the firebase studio documentation. I logged in using the firebase login --no-localhost command in the terminal. Not sure If I am missing something

I am trying to connect to the firebase studio prototyper to the firebase MCP but have been having some trouble. DO you have any insights? Thanks in advance.

I have been getting the same error in the past 48 hours or so, it looks like they updated the server version or something, if i figure it out i will post it here.

1 Like

Sounds good thanks mate! I was trying for quite some time and couldn’t figure out what was wrong. Good to know its not just me.