Blueprint

The Blueprint MCP Server integrates with Arcade’s ecosystem to generate visual diagrams from codebases and system documentation using Nano Banana Pro.

It creates architecture diagrams, flowcharts, and system visualizations that help you understand complex code structures and technical processes.

Features

• 🧩 Generates architecture diagrams from code analysis
• 🔄 Creates sequence diagrams for API flows
• 📊 Produces data flow diagrams for ETL pipelines
• 🔗 Integrates with Arcade’s tool ecosystem
• ⚡ Uses Nano Banana Pro for fast diagram generation
• 🖼️ Exports diagrams as base64-encoded PNG files
• 🎯 Supports enterprise architecture scenarios
• 📚 Creates educational learning cards for frameworks

Installation

1. This MCP server requires a connection to the Arcade platform and a Google AI Studio API key.

• Arcade: Sign up at arcade.dev.
• Google AI Studio: Generate an API key at aistudio.google.com.

2. Set up a Python virtual environment to isolate dependencies.

# Create virtual environment
python3 -m venv venv
# Activate the environment
# On macOS/Linux:
source venv/bin/activate
# On Windows:
venv\Scripts\activate
# Install the Arcade CLI
pip install arcade-mcp

3. Authenticate the CLI and securely store your Google API key.

# Log in to your Arcade account
arcade-mcp login
# Set the Google API key as a secret
arcade-mcp secret set GOOGLE_API_KEY="your_api_key_here"

4. Deploy the MCP server code to the Arcade runtime.

# Navigate to the server directory
cd architect_mcp
# Deploy the server
arcade-mcp deploy

5. The deployed MCP server needs to be exposed via a Gateway to be accessible by clients like Cursor.

1. Navigate to the Arcade Dashboard.
2. Select Gateways and click Create Gateway.
3. Locate your deployed architect_mcp server in the list and add it to this gateway.
4. Copy the resulting Gateway URL.

6. Configure your AI assistant like Cursor:

• Open the Cursor app and go to Settings > MCP.
• Click Add new MCP server.
• Select “SSE” (Server-Sent Events) if prompted, or simply paste the Arcade Gateway URL you copied in the previous step.
• Restart Cursor to load the tools.

Tool Usage and Workflow

Available Tools:

• start_diagram_job: Initiates the request.
• check_job_status: Polls for completion.
• download_diagram: Retrieves the final image.

Step-by-Step Execution:

1. Call start_diagram_job with your prompt.

• Example Prompt: “Analyze the authentication module in src/auth/ and create an architecture diagram.”
• Result: Returns a job_id.

2. The Nano Banana Pro model takes approximately 30 seconds to generate the diagram.

3. Call check_job_status using the job_id from step 1.

• Repeat this until the status returns “Complete”.

4. Call download_diagram with the job_id.

• Result: Returns a base64 encoded PNG string. Your MCP client (Cursor) should decode this and save it to your workspace automatically.

FAQs

Q: Why doesn’t the diagram appear instantly?
A: The diagram generation is computationally intensive. The server uses an asynchronous job queue. You must wait roughly 30 seconds after starting a job before the result is ready for download.

Q: Can I use this with code that isn’t in a public repository?
A: Yes. When you use this in an editor like Cursor, the editor sends the relevant code context from your local files to the MCP server as part of the prompt. The server processes this context to generate the diagram.

Q: What happens if I lose my Job ID?
A: You will need to restart the generation process. The Job ID is the only key to retrieving the specific diagram you requested.

Q: Can I customize the colors or style of the diagram?
A: You can influence the style via the prompt. For example, you can specify “Use technical whiteboard style, muted colors (gray, light blue)” in your initial request to start_diagram_job.