Assistente para Desenvolvedores Google
Siva-Subramaniam-DS/grounded-developer-assistant
Chat inteligente com docs oficiais Google integradas via MCP
Ver no GitHubMCP Server que oferece uma interface Streamlit interativa para consultas técnicas sobre Google Cloud, Firebase, Android e Maps, com respostas ancoradas em documentação oficial. Utiliza Model Context Protocol e Gemini API para reduzir alucinações e fornecer fontes verificáveis com cada resposta.
Instalação
npx convez-ai Siva-Subramaniam-DS/grounded-developer-assistant1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Consultas técnicas sobre Google Cloud Platform com fontes oficiais
- Suporte a desenvolvedores Android com documentação integrada
- Pesquisa em documentação Firebase com rastreamento de fontes
- Respostas RAG grounded em APIs e serviços Google
Veredito

Solução robusta e bem arquitetada para desenvolvedores que precisam de respostas confiáveis sobre ecossistema Google. A implementação de dois modos de grounding (RAG + Managed Answer API) é sofisticada, mas exige configuração OAuth2 e conhecimento técnico significativo.
README
A tradução para o português deste README ainda está na fila.
⚡ Grounded Developer Assistant
A premium, interactive Streamlit chat interface designed to act as a Google Developer Assistant. By leveraging the Model Context Protocol (MCP) and Google's Developer Knowledge API, the assistant provides real-time technical answers grounded strictly in official documentation for Google Cloud, Firebase, Android, and Google Maps.
📖 Table of Contents
- Key Features
- Architecture Overview
- Grounding Modes
- Prerequisites
- Installation
- Configuration & Usage
- Testing and Helper Scripts
- Project Tags & Metadata
✨ Key Features
- ⚡ Premium User Experience: Built with custom styled components, a vibrant header banner utilizing HSL/linear-gradients, custom interactive source cards, and dynamic element rendering.
- 📚 Fully Grounded AI Responses: Answers are strictly backed by Google's official documentation datasets, significantly reducing hallucinations.
- 🛠️ Flexible Grounding Engines: Supports both custom standard Retrieval-Augmented Generation (RAG) and Google's Managed Answer API.
- 🔐 Seamless Authentication: Features automatic OAuth2 access token detection using the local Google Cloud SDK (
gcloudCLI) as well as manual credential entry. - 💬 Streaming Conversations: Interactive chatbot with persistent session memory and real-time word-by-word streaming generation.
- 🌐 Clickable Sources & References: Shows expandable grounded sources containing text snippets and direct web links to official documentation.
🖥️ Architecture Overview
graph TD
User([User Prompt]) --> Streamlit[Streamlit UI app.py]
Streamlit --> Auth{OAuth2 Token & API Key}
Auth -->|Valid| ModeSelect{Grounding Mode?}
ModeSelect -->|Standard RAG| Search[MCP search_documents]
Search -->|Retrieved Docs| Context[Construct Context]
Context --> Gemini[Gemini Client / API]
Gemini -->|Streamed Response| Output[UI Chat Window]
ModeSelect -->|Managed Answer| AnswerAPI[MCP answer_query]
AnswerAPI -->|JSON Response| Parser[Parse Answer & Refs]
Parser -->|Format output| Output
🔍 Grounding Modes
The application supports two distinct grounding workflows configured in the sidebar settings:
1. Standard RAG (Search + Gemini)
- Retrieve: Queries the Developer Knowledge MCP Server (
search_documentstool) with the user prompt. - Ground: Formulates a strict system prompt containing the search results/snippets.
- Synthesize: Calls Gemini API (using the new
google-genaiSDK) to draft a response based only on the retrieved context.
2. Managed Answer API (answer_query)
- Direct Execution: Directly calls the Google Developer Knowledge MCP
answer_querytool. - Parse & Render: The backend returns pre-compiled answers along with corresponding official document references, which are structured and styled as rich interactive cards.
📋 Prerequisites
Before running the application, make sure you have:
- Python 3.10+ installed.
- Google Cloud SDK (
gcloudCLI) installed and authenticated:
Note: The app will rungcloud auth login gcloud auth application-default logingcloud auth print-access-tokenin the background to automatically retrieve your OAuth2 credentials. - An active Google Cloud project with the Developer Knowledge API enabled.
- A Gemini API Key from Google AI Studio.
🚀 Installation
- Clone or copy this repository to your local workspace.
- Install the required Python dependencies:
pip install streamlit google-genai requests
⚙️ Configuration & Usage
- Start the Streamlit application:
streamlit run app.py - Open the browser window at
http://localhost:8501. - In the sidebar settings:
- Gemini API Key: Input your API key (defaults to your environment's
GEMINI_API_KEY). - GCP Project ID: Input your target Google Cloud Project ID (e.g.,
geogreenai). - OAuth2 Token: Automatically detected if your
gcloudCLI is logged in. Otherwise, paste a manual token. - Model Settings: Choose from
gemini-2.5-flash,gemini-1.5-flash, orgemini-1.5-proand adjust the creativityTemperature. - Grounding Mode: Choose between Standard RAG and Managed Answer API.
- Gemini API Key: Input your API key (defaults to your environment's
- Start chatting with the grounded developer assistant!
🧪 Testing and Helper Scripts
The codebase comes equipped with several lightweight test scripts to verify API connections and test components in isolation:
- test_gemini.py: Verifies the
google-genaiSDK connectivity and simple text generation. - test_mcp.py: Tests the connection to the Developer Knowledge MCP Endpoint and lists available tools.
- test_mcp_oauth.py: Tests searching documents via OAuth2 authorization.
- test_mcp_search.py: Verifies document searching functionality using raw API keys.
- test_mcp_answer.py: Verifies target answer generation using the managed
answer_querytool. - test_parser.py: A local unit test for regex/parsing rules on the raw protocol response blocks.
🏷️ Project Tags & Metadata
- Category: AI Assistant / RAG / Developer Tools
- Frameworks: Streamlit, Google GenAI SDK, Model Context Protocol (MCP)
- Target Audiences: Cloud Engineers, Firebase Developers, Android Engineers, Google Maps Developers
- Languages: Python
- Design Keywords: HSL Gradients, Dark Theme, Responsive UI, Streamlit custom CSS