Firebase Docs MCP Server Setup

Directory Layout

docs-mcp

This corresponds to the indexer for Firebaes docs. This is a Go project that goes and indexes the Firebase documents contained within the listed filepaths.

docs-mcp-server

This is the model context protocol server that serves content over a stdio transport.

genkit-mcp-tester

This is a genkit implementation of an MCP client to test using the docs-mcp-server.

How to use

Start with indexing

1. Set the API Key. We are using the Gemini embedding model for the documents so getting an API key from AI Studio is required. To set the API key, call export genaikey="APIKEY" in your terminal

1. Ensure that the output directory is empty. We are writing files to your home
directory in a folder called .indexResp. As go fetches documents from the
Firebase documentation site, it writes the files to disk in markdown format and
also indexes them in a SQL lite database in this directory. If indexing fails,
it performes a retry strategy to reindex the documents into a markdown format.

1. From the docs-mcp folder, call go run . This will start the indexing
process on the files listed near line 291 in the main.go file.

Test the indexer

1. Set the API Key. We are using the Gemini embedding model for the documents so getting an API key from AI Studio is required. To set the API key, call export genaikey="APIKEY" in your terminal

1. Switch into the docs-mcp-server folder.

1. Copy the indexed database to the local docs-mcp-server folder. This can be
done by calling cp $HOME/.indexResp/db.sqlite .

1. Install the dependencies and build the project. npm ci and then
npm run build. Once the project is built, you can then test the project by
calling npm run build && npx @modelcontextprotocol/inspector node build/index.js.
This starts the inspector and should print a URL for you to view the STDIO server with.

1. Click on Connect in the inspector view, and then click on tools -> List Tools
-> find-firebase-doc and then type in for your request that you would want to
use. NOTE: The author has had trouble using the terminal built into VSCode
for running this step, so if you run into a similar issue, try the system
terminal.

Use Genkit for testing

1. Set the API key in the code by changing this line in embedding.ts from : const genAiKey = process.env.genaikey || ""; to const genAiKey = process.env.genaikey || "MYAPIKEY";

1. Switch into the genkit-mcp-tester directory.

1. Copy the indexed database to the local genkit-mcp-tester folder. This can be
done by calling cp $HOME/.indexResp/db.sqlite .

1. Install the dependencies and build the project. npm ci and then
npm run build. Once the project is built, you can then test the project by
calling npx genkit start -- npx tsx --watch src/index.ts.
This starts the Genkit DevUI where you can interact with the flow and tool
directly. Open the DevUI, generally http://localhost:4000
and visit the Tools -> find-firebase-doc/find-firebase-doc tool and make a
request here. You can see that the request is then returning the results we see
in the modelcontextprotocol/inspector.