This site is also available in:
In today’s article, we look at how banks can implement a context-based communication platform based on the MCP principle. We summarize key requirements, market solutions and open source frameworks and compare them with LeapXpert (Swisscom) and Agentforce (Salesforce) as examples.
1 Introduction: Why MCP and Agent AI are crucial for banks
Banks are under pressure to establish AI-driven, context-based communication platforms that work seamlessly across countries and channels. With Model Context Protocol (MCP) and Agent AI, existing APIs and integrations can be transformed into intelligent, agent-enabled services for next-best-action, automated summaries and real-time consent-based data processing.
2 Central requirements for an MCP-capable communication platform
Agent-based communication & Agent AI
AI agents (e.g. chatbots, virtual assistants) use context data for precise answers and actions.
Context-controlled workflows
Dynamic process control via DMN decision rules (e.g. channel selection, offer creation).
API orientation & MCP
Exposure of all backend systems via REST/GraphQL APIs as MCP server.
Cross-channel collaboration
Standardized message and agent control via chat apps, e-mail, SMS, voice, video.
AI functions for banks
Next-best-action, automatic summaries, sentiment analysis.
Consent management
Strict adherence to consent in every customer interaction.
3. market overview: Commercial MCP & Agent AI platforms
| Solution | Core functions | Banking focus |
|---|---|---|
| MuleSoft MCP Support (Beta) | APIs & integrations as MCP server, standardized agent-to-tool protocol, secure scaling | Enables AI agents next-best-action & real-time context |
| LeapXpert | Compliance chat in messaging apps, archiving, consent handling | Specialized chat compliance |
| Symphony | Secure multi-channel communication, bot/API connection, AI summaries | Trading & Wealth Management |
| Unblu | Omnichannel (chat, video, co-browsing), secure messenger, AI co-pilot | Banks & insurers |
| Qwil | Cloud messenger with end-to-end encryption, file/screen sharing | Banks, wealth management, law firms |
| Infobip (CPaaS) | SMS, WhatsApp, email, voice, chatbot builder, campaign workflows | Transaction notifications across all sectors |
| Agentforce (Salesforce) | Unified CRM platform, autonomous workflow management | Sales and service teams in banks |
4. open source frameworks for MCP & Agent AI
| Framework | Functions | Banking use |
|---|---|---|
| Rasa | NLU, context-driven stories, multi-channel connectors, on-prem option | Chatbots, virtual assistants with agent AI function |
| CIBSeven | BPMN engine, DMN decision rules, banking connectors, on-prem performance | Orchestration of complex workflows & next-best-action |
| MuleSoft Anypoint | MCP connector, API management, integration templates | Seamless exposure of all systems as MCP servers for AI agents |
5. example architecture: MCP & Agent AI in banking operations
Data Hub / Data Lake
Saves customer profiles, consent records, event logs.
Provides context data (last contact, open opportunities) via REST/GraphQL.
CIBSeven BPMN & DMN
Orchestrates “customer interaction” and “next-best-action” processes.
Service APIs
Consent service: Checks consent before interaction.
CRM service: Writes after-call summaries & NBX recommendations to the CRM.
NBX engine: Machine learning-based action suggestions.
MuleSoft MCP Server: Exposes MuleSoft APIs & integrations as MCP for Agent AI.
Messaging gateway
Delivery via LeapXpert, Infobip, Unblu, Qwil.
Rasa Agent AI
Uses Rasa NLU/Stories for initial contact, forwards complex requests to agents.
Monitoring & Audit
Complete logging of all process instances, decisions, API calls in CIBSeven Cockpit.
6 Conclusion: Flexibility & future-proofing thanks to MCP & Agent AI
With Model Context Protocol (MCP) and Agent AI, banks are creating a modular, context-driven and AI-enabled communication platform. The building block approach – from MuleSoft MCP and LeapXpert to Rasa and CIBSeven – allows next-best-action, compliance and omnichannel workflows to be optimally linked. So financial institutions are fit for the AI-native future!
