Tanzu Platform Chat: AI Chat Client for Cloud Foundry

Overview

Tanzu Platform Chat (cf-mcp-client) is a Spring chatbot application that can be deployed to Cloud Foundry and consume platform AI services. It's built with Spring AI and works with LLMs, Vector Databases, and Model Context Protocol Agents.

Prerequisites

• Java 21 or higher
  • e.g. using sdkman sdk install java 21.0.7-oracle
• Maven 3.8+
  • e.g. using sdkman sdk install maven
• Access to a Cloud Foundry Foundation with the GenAI tile or other LLM services
• Developer access to your Cloud Foundry environment

Deploying to Cloud Foundry

Preparing the Application

1. Create a directory for the application and navigate to it:

mkdir tanzu-platform-chat
cd tanzu-platform-chat

2. Download the latest JAR file and manifest.yml from the Releases page into this directory

3. Push the application to Cloud Foundry from the directory containing the downloaded files:

cf push

Authentication

The application requires authentication to access. There are two authentication methods available: access code and SSO. Both can be used simultaneously.

Access Code Authentication

By default, the application is protected with an access code. Users enter the code on the login page to gain access.

If APP_AUTH_SECRET is not set, the access code defaults to changeme.

To override the default access code, you have two options:

Option 1: Set as an environment variable

cf set-env ai-tool-chat APP_AUTH_SECRET my-secret-code
cf restart ai-tool-chat

Option 2: Add to manifest.yml

Add the environment variable to your manifest.yml:

env:
  JBP_CONFIG_OPEN_JDK_JRE: '{ jre: { version: 21.+ } }'
  APP_AUTH_SECRET: ((APP_AUTH_SECRET))

Then provide the variable at push time:

cf push --var APP_AUTH_SECRET=my-secret-code

SSO Authentication (Optional)

Single Sign-On is automatically enabled when a p-identity service is bound to the application. The login page will display a "Sign in with SSO" button alongside the access code form.

1. Create a SSO service instance:

cf create-service p-identity uaa my-sso

2. Bind the service to your application:

cf bind-service ai-tool-chat my-sso
cf restart ai-tool-chat

The SSO provider is auto-detected from the Cloud Foundry service binding — no additional configuration is required.

Binding to Large Language Models (LLM's)

1. Create a service instance that provides chat LLM capabilities:

cf create-service genai [plan-name] chat-llm

2. Bind the service to your application:

cf bind-service ai-tool-chat chat-llm

3. Restart your application to apply the binding:

cf restart ai-tool-chat

Now your chatbot will use the LLM to respond to chat requests.

Binding to Vector Databases

1. Create a service instance that provides embedding LLM capabilities

cf create-service genai [plan-name] embeddings-llm 

2. Create a Postgres service instance to use as a vector database

cf create-service postgres on-demand-postgres-db vector-db

3. Bind the services to your application

cf bind-service ai-tool-chat embeddings-llm 
cf bind-service ai-tool-chat vector-db

4. Restart your application to apply the binding:

cf restart ai-tool-chat

5. Click on the document tool on the right-side of the screen, and upload a .PDF File

Now your chatbot will respond to queries about the uploaded document

Binding to MCP Agents

Model Context Protocol (MCP) servers are lightweight programs that expose specific capabilities to AI models through a standardized interface. These servers act as bridges between LLMs and external tools, data sources, or services, allowing your AI application to perform actions like searching databases, accessing files, or calling external APIs without complex custom integrations.

Method 1: Using User-provided Services

SSE Protocol (Server-Sent Events)

1. Create a user-provided service for an SSE-based MCP server using the mcpSseURL tag:

cf cups mcp-server-sse -p '{"uri":"https://your-sse-mcp-server.example.com"}' -t "mcpSseURL"

2. Bind the MCP service to your application:

cf bind-service ai-tool-chat mcp-server-sse

Streamable HTTP Protocol

1. Create a user-provided service for a Streamable HTTP-based MCP server using the mcpStreamableURL tag:

cf cups mcp-server-streamable -p '{"uri":"https://your-streamable-mcp-server.example.com"}' -t "mcpStreamableURL"

2. Bind the MCP service to your application:

cf bind-service ai-tool-chat mcp-server-streamable

Method 2: Using Service Publisher provided Services

A common use case for the Service Publisher tile is to publish MCP servers and make them available on the Tanzu Platform service marketplace. The service binding will include an API Key and URI allowing for more secure provisioning of access to MCP servers. The application will add the API key to requests to the MCP server.

Published services will appear on the marketplace with their own service broker and selection of plans.

SSE Protocol (Server-Sent Events)

1. Create a service from a published service broker for an SSE-based MCP server using the mcpSseURL tag:

cf create-service your-published-service service-plan mcp-server-sse -t "mcpSseURL"

2. Bind the MCP service to your application:

cf bind-service ai-tool-chat mcp-server-sse

Streamable HTTP Protocol

1. Create a service from a published service broker for a Streamable HTTP-based MCP server using the mcpStreamableURL tag:

cf create-service your-published-service service-plan mcp-server-streamable -t "mcpStreamableURL"

2. Bind the MCP service to your application:

cf bind-service ai-tool-chat mcp-server-streamable

Complete the Setup

3. Restart your application to apply the bindings:

cf restart ai-tool-chat

Your chatbot will now register with the MCP agents, and the LLM will be able to invoke the agents' capabilities when responding to chat requests. The application supports both SSE and Streamable HTTP protocols simultaneously.

Binding to A2A Agents

Agent2Agent (A2A) is a protocol that enables communication between independent AI agent systems. Unlike MCP servers which provide tools that your LLM can invoke, A2A agents are independent AI systems that can process messages and return their own intelligent responses. Think of MCP as giving your chatbot tools to use, while A2A lets your chatbot consult with other specialized AI agents.

Key Differences: MCP vs A2A

• MCP Servers: Provide tools and data sources that your LLM invokes as part of generating a response
• A2A Agents: Independent AI agents that you can send messages to directly and receive complete responses from

Binding an A2A Agent

1. Create a user-provided service for an A2A agent using the a2a tag:

cf cups a2a-agent -p '{"uri":"https://your-a2a-agent.example.com/.well-known/agent.json"}' -t "a2a"

The URI should point to the agent's Agent Card (a JSON descriptor at /.well-known/agent.json).

2. Bind the A2A service to your application:

cf bind-service ai-tool-chat a2a-agent

3. Restart your application to apply the binding:

cf restart ai-tool-chat

Using A2A Agents

After binding A2A agents:

1. Click the Agents button (🤖) on the navigation rail
2. The Agents panel shows all connected A2A agents with their:
   • Name, description, and version
   • Health status (healthy/unhealthy)
   • Capabilities (streaming, push notifications, state history)
3. Click "Send Message" on any healthy agent
4. Type your message in the dialog and click "Send"
5. The agent's response appears in the chat with a distinct visual style

Agent messages are displayed with:

• A robot icon (🤖) and agent name header
• Tertiary color scheme (different from your LLM's responses)
• Clear attribution showing which agent responded

You can bind multiple A2A agents simultaneously, each providing specialized capabilities. The Agents panel displays health status for each agent, and the navigation button shows a status indicator (green/orange/red) based on overall agent health.

Using a Vector Store for Conversation Memory

If you are bound to a vector database and an embedding model, then your chat memory will persist across application restarts and scaling.

1. Follow the instructions above in Binding to Vector Databases

Disclaimer

This repository is an unofficial project provided “as is.” It is not supported or endorsed by any organization, and no warranty or guarantee of functionality is provided. Use at your own discretion.