The integration of backend systems into a unified agent desktop that couples computer and telephony integration, that is scalable, secure, reliable, fault-tolerant, with role based access permissions, that performs well at scale has always been challenging.  The challenge is significant enough, that even some of the worlds largest and best integrated contact centers may only have a small handful of deeply integrated core applications available to the agent, relying instead of 'swivel chair integration' when agents need access to and from other systems.  Agents become very good at copying and pasting information from System A to System B, excellent at enterprise search and using a multitude of system user interfaces, all at the same time.

Moving the enterprise quickly to A.I bots does not relieve the enterprise of the integration burden, indeed, in some ways it makes it worse!  The cost of deploying an enterprise specific LLM, trained, shaped and honed to meet the enterprise needs are extremely high, as is the risks associated with the guardrails and controls that would need to be in place to ensure your AI bot stayed true to its stated mission and purpose reliably, securely and assuredly.  These are subject though for a different article.

This is why we are all seeing so many FAQ style AI Bots and chat assistants all around us.  They do not need to be integrated with backend systems of record.  They use instead a much weaker integration approach called Retrieval Augmented Generation, or RAG for short.  In RAG, the enterprise feeds the A.I engine of choice with enterprise specific documentation and FAQ style questions and answers.  Most mainstream LLM's are capable of leveraging attached documents under the RAG methodology to become more aligned to the enterprise that they are representing.

But, the big thing that RAG leaves behind, is the ability to do something useful.  It is naturally, great for the enterprise that a routine enquiry about the hours of operation can be answered by a BOT.  That's great, it just prevented a call.  But, the true deep essence of the modern contact center is not the handling of high volumes of low value transactions, moreover the handling of high volumes of complex interactions that require access to and from backend systems of record.  It's here where RAG falls flat, and fails to deliver.  Sure, it will shave off some basic FAQ style enquiries, but it won't deal with the tasks, activities and deliverables required for good customer experience management.

By its very nature, delivering a highly optimized, efficient and effective human to human conversation that bridges enterprise systems of record is challenging.  It it has been difficult, too expensive, too cumbersome or otherwise challenging for the enterprise to deal with when considered in a traditional contact center the use of an AI Bot to replace a human agent is no different.  In fact, its worse!  A human agent has significantly more heuristic intelligence to spot attacks, fraud, lies, innuendo, or bad actors. A human agent will understand exactly how to find the key information, or exactly how to perform a complex piece of workflow to achieve a given task, for example, a customer wanting to update their home address on record and at the same time arrange for the move of the product or service subscription to that new address on a given date may require a CRM update, a billing system update, a new task on the delivery/fulfillment team, a new order with external suppliers and so on.  A general AI BOT fueled by basic enterprise FAQ documentation will not achieve this.

To do this, and to do it well, at scale, requires agentic A.I.  Agentic A.I benefits from the ability to act, make decisions and take actions to achieve its goals without much or any human intervention.

To build agentic A.I bots for voice and digital services that are integrated into your overall customer and employee experience you will need to integrate your AI framework with your enterprise systems.  The Model Context Protocol has been agreed on and adopted by most Ai vendors including OpenAI (ChatGPT), Google, Microsoft, Amazon, Cloudflare and others as the agreed-upon open standard in the field of A.I. for connecting A.I engines and services to backend systems of any type.  First proposed by Anthropic (Claude) in November 2024 it is gaining momentum and is becoming more widely adopted.  Just as the IVR industry moved from proprietary service creation tools and languages towards VoiceXML and CC-XML (Call Control) the same has happened in A.I.  MCP is the future and it is here today.

How does it work?

Your AI Host service calls an MCP Agent as if it were an intrinsic tool.  When the AI Host recognises that it needs information from a backend system, or wants to pass a request to another system, it sends its request to a dedicated, or multiple MCP Servers.  The MCP servers act as adapters for the data sources. The MCP server authenticates the request, determines the intent, and acts as a router of the request, sending it in the correct format to the target system.  It then formulates a response and sends it back to the AI Host in a predefined format.  The AI Host consumes the result and uses its own system prompt and guardrails to handle the returned data and offering it up to the end user.

This seems simple in principle.  But, it has inherent complexity.  Your MCP server must track all sessions, handle authentication, be performant at scale, be reliable and be capable of describing its capabilities and registry of functions to the A.I. agent, for example, query database, call a CRM API, retrieve a document and so on.

In short, as versed as QVCCS in enterprise integration for traditional contact centers, we are also specialists in MCP as an integration model to the same backend systems we have been integrating for over 40 years.

If you are contemplating the consumption of AI Bots for Voice and Digital and need assistance with MCP as a vehicle to achieve your automation then please do not hesitate to contact us.