Download to read offline
AI Innovations with MCP and Curie Tech | Kolkata MuleSoft Meetup #22
- 1. July 12, 2025 KolkataMuleSoft Meetup
- 2. Priya Shaw JyotiChaubey Indranil Paul Moderators / Organisers Suman Chatterjee Brojeswar Roy
- 3.
- 4. ● What isModel Context Protocol (MCP)? ● Evolution of MCP ● How it solves problem? ● MuleSoft MCP Support Overview ● MCP Demo ● Q&A Agenda
- 5. MCP is anopen protocol that standardizes how application provides context to LLMs. Think of MCP like a USB C port for AI application. What is Model Context Protocol (MCP)?
- 6. Evolution of ModelContext Protocol (MCP) Phase 1 Generative AI (LLM) Phase 2 LLM + context Phase 3 MCP (Fixing the integration mess)
- 7. Evolution of ModelContext Protocol (MCP)
- 8. Difficult to scale AIAgent (powered by Claude) Customer Data Inventory Data ● Descriptions ● Tools ● Resources ● Descriptions ● Tools ● Resources ● Descriptions ● Tools ● Resources Internal Data AI Agent (powered by OpenAI) AI Agent (in Cursor)
- 9. Transform Any APIInto An Agent-Ready Asset With Mulesoft MCP Support Customer Data Inventory Data ● Descriptions ● Tools ● Resources MC P Internal Data AI Agent (powered by OpenAI) AI Agent (powered by Cursor AI) AI Agent (powered by Claude)
- 10. MCP Architecture: Simpleand Powerful MCP follows a client-server architecture where: ● Hosts are LLM applications (like Claude Desktop or IDEs) that initiate connections ● Clients maintain 1:1 connections with servers, inside the host application ● Servers provide context, tools, and prompts to clients Source - https://modelcontextprotocol.io/docs/concepts/architecture MCP Protocol MCP Protocol Code REPO (Github) Database (Postgre SQL) Managed by service provider
- 11. MCP Architecture: Howit works internally Source - https://modelcontextprotocol.io/docs/concepts/architecture
- 12. Bring MCP ToEvery API Expose any API as an MCP server in minutes using pre- built operations Build Context-Rich AI Applications Allow models and agents to expand their knowledge from any source system Create Secure And Scalable AI Ecosystems Secure and standardize communication with AI agents and models through a unified platform MuleSoft MCP Support GA | Now Transform any app or API into an agent-ready asset Model Context Protocol Connec… Model Context Protocol ●Descriptions ●Tools ●Resources M C P
- 13.
- 14.
- 15. The Composable ArchitectureUnlocks The Enterprise Netsuite Your Backend Systems Your User Experiences Customer API Customer Mobile System Layer Experience Layer Process Layer CRM Pipeline Dashboards B2B Partner Portal Orders API Finance API Product API RPA Bots LLM API OMS API Stripe Legacy System Salesforce LLM 2 Homegrown OMS Documents Document Processor SAP API Employee API Salesforce Flow Inventory API SAP Vector Database Fulfillment API SAP AI Chain
- 16. Actionability Is TheKey To Agentic Success Netsuite Your Backend Systems Customer API System Layer Experience Layer Process Layer Your User Experiences Customer Mobile CRM Pipeline Dashboards B2B Partner Portal Orders API Finance API Product API RPA Bots LLM API OMS API Stripe Legacy Salesforce LLM 2 Homegrown Documents Document Processor SAP API Employee API Salesforce Flow Inventory API SAP Vector Database Fulfillment API AI Chain SAP Action Layer Model Context Protocol (MCP) Topic Center Your Digital Workforce Field Agent Sales Agent Service Agent Fulfillment Agent
- 17. Transform Any APIInto An Agent-Ready Asset With MuleSoft Internal Data Customer Data Inventory Data Internal Data Customer Data Inventory Data MuleSoft for Agentforce MuleSoft for Third-Party Agents Topic Center MuleSoft MCP Support Product Agent (powered by Agentforce) Product Agent (powered by OpenAI)
- 18.
- 19.
Editor's Notes
- #6 Anthropic’s MCP Announcement: https://www.anthropic.com/news/model-context-protocol
- #7 Anthropic’s MCP Announcement: https://www.anthropic.com/news/model-context-protocol
- #8 Talk Track Scalability of AI applications are rooted in the ability to easily provide models and agents with your organizations business context. Agents and models require instructions on when to use each data source, along with additional context and metadata to help use the data appropriately. However, with over 1000+ applications in the average enterprise, doing this manually for each data source is not sustainable. For example: An AI agent powered by Claude requires customer data to be more effective at answering relevant customer questions. <click> A developer would need to build a new integration from this agent to their customer data source (Salesforce) and embed instructions, tools, and resources manually so the agent can properly use the customer data. <click> What if we wanted to further extend the agents reach to Inventory? The same process would need to be repeated. <click> This would happen again with every additional data source, and if you were to add a new agent in the mix, the process would need to be repeated for all 3 sources. This slows the pace of innovation at your organization. <click> And what happens when you add more agents? It becomes even more complex.
- #9 Talk Track Let’s see how the scenario above changes with MuleSoft MCP Support. <Click> First we expose our data sources – Salesforce, SAP, internal databases etc. – once as MCP Servers using the MCP Connector. Here we define the available context, instructions, and metadata for each source within MuleSoft, conforming to the MCP standard. Our AI agents will act as MCP Clients. They learn to ask for context using the standardized MCP format. They no longer need custom-built pipelines for every single data source. <Click> So, when you want your Claude agent to access Inventory data? You just expose Inventory as an MCP Server through MuleSoft. The agent doesn't need to change. It already knows how to speak MCP, and now Inventory is available on the 'network' of context providers (available MCP Servers) <Click> And when you add that second agent? As long as it's built to be an MCP Client, it can immediately access both Salesforce and Inventory context through the MCP Servers you've already configured in MuleSoft. No redundant integration work for the new agent. This breaks the N x M integration nightmare. Instead of point-to-point connections, you have agents connecting to a standardized context fabric, and data sources plugging into that same fabric, all orchestrated by MuleSoft. Furthermore, MuleSoft provides the essential governance, security, and observability around these interactions. You can manage access, monitor usage, and ensure your AI is using data securely and appropriately, all through Anypoint Platform. By adopting this standardized Model Context Protocol approach with MuleSoft, you drastically reduce integration friction. You empower developers to rapidly and reliably connect AI to the business context it needs, accelerating the pace of innovation and unlocking the true scaling potential of AI within your organization.
- #12 Context: Scalability of AI applications are rooted in the ability to easily provide models and agents with your organizations business context. Agents and models require instructions on when to use each data source, along with additional context and metadata to help use the data appropriately. However, with over 1000+ applications in the average enterprise, doing this manually for each data source is not sustainable. Talk Track: To solve this, Anthropic released an open-source protocol called MCP (Model Context Protocol), which is a new standard for connecting AI assistants and models to the systems where data lives, including content repositories, business tools, and development environments. Its aim is to help frontier models produce better, more relevant responses. MuleSoft is embracing this new technology by releasing MCP support on Anypoint Platform. This allows organizations to easily adopt MCP through a pre-built connector that can be used in new or existing Mule apps, making them compatible with other MCP-supported agents and models. Bring MCP To Every API MuleSoft’s implementation of MCP is designed for flexibility, speed, and scale. With just a few clicks, developers can transform their APIs and Mule apps into MCP servers — no complex setup or manual configuration required. Build Context-Rich AI Applications By supporting MCP, MuleSoft is empowering organizations to build AI agents and models that are able to seamlessly access and utilize data from any connected system for richer context. Create Secure And Scalable AI Ecosystems MuleSoft enables secure interactions between applications and AI technologies through a unified integration and API management platform, while also helping conform to the new industry standard for exposing information to AI agents (MCP). This allows teams to easily work with any AI agents while ensuring security and governance at all times.
- #17 Talk Track
- #19 Thank you.