It’s live! Access exclusive 2025 live chat benchmarks & see how your team stacks up.
Get the data
AI agents promise to transform business operations. But there’s a dirty secret the industry won’t tell you: most “intelligent” systems can’t easily connect to the tools they claim to enhance.
The problem isn’t capability; it’s connectivity. Every AI integration becomes a custom engineering project that takes months to build, costs six figures, and breaks whenever APIs change.
The Model Context Protocol (MCP) is Anthropic’s attempt to solve this mess. But unlike the rosy demos you’ve seen, the reality of MCP adoption is more complex. Early enterprise implementations reveal both breakthrough potential and serious limitations that vendors prefer not to discuss.
Here’s what’s actually happening with MCP in the real world.
The Model Context Protocol (MCP) is an open standard developed by Anthropic (the company behind Claude) in November 2024 that enables AI applications to connect with external tools, databases, and services through a unified interface.
Instead of building custom integrations for each data source, MCP provides a standardized communication protocol that allows any MCP-compatible AI system to interact with any MCP server.
The protocol works through a client-server architecture where AI applications act as clients that send standardized requests to MCP servers.
These servers translate the requests into native API calls for specific systems like Salesforce, Google Drive, or internal databases, then return responses in a format the AI can understand.
This approach transforms the traditional M×N integration problem; where M AI applications connecting to N data sources require M×N custom integrations, into a simpler M+N model where each system only needs one MCP implementation.
MCP defines three core capabilities that AI agents can access:
The protocol uses JSON-RPC 2.0 for message formatting and supports both local connections (via STDIO) and remote connections (via HTTP with Server-Sent Events).
Major technology companies have already adopted MCP across their platforms, and more and more businesses across multiple sectors are following suit (more on that below).
However, enterprise deployment reveals significant challenges. Most publicly available MCP servers are designed for development use cases rather than production environments.
They often lack robust authentication, multi-tenancy support, and the error handling required for enterprise-scale operations. Organizations implementing MCP successfully typically build additional infrastructure layers to address these limitations while maintaining the protocol’s standardized interface.
MCP operates on a straightforward client-server model where AI applications function as clients that communicate with MCP servers hosting specific capabilities.
The client initiates connections, sends requests, and processes responses, while servers expose their functionality through standardized interfaces that any MCP-compatible client can understand.
When an AI application needs to access external data or tools, it connects to the appropriate MCP server and performs a capability negotiation handshake.
During this process, the server announces what resources, tools, and prompts it offers, along with detailed descriptions of each capability. This allows the AI to understand what actions it can perform without requiring prior knowledge of the specific system.
A typical MCP workflow follows these steps:
For example, asking “Why was I charged $50 more this month?” might trigger requests to a CRM server for data and an analytics server for reporting functions, with the AI combining both responses into a complete transaction summary.
MCP organizes capabilities into three distinct categories that cover different interaction patterns. Resources represent read-only data access, such as retrieving customer records, reading documents, or querying databases.
These operations don’t modify system state but provide the contextual information AI agents need to understand current situations.
Tools enable AI agents to take actions with side effects, like creating support tickets, sending emails, or updating records. Each tool defines its required parameters, expected behavior, and potential outcomes, allowing AI systems to reason about which tools to use for specific tasks.
Prompts provide templated interaction patterns for complex, multi-step workflows. Rather than requiring AI agents to figure out optimal tool sequences through trial and error, prompts offer curated approaches that combine multiple resources and tools to accomplish common business objectives efficiently.
MCP plays a critical role for companies deploying AI-powered customer support solutions:
Customer service organizations operate across multiple disconnected platforms that create operational silos and limit agent effectiveness. A typical support team uses separate systems for ticketing, knowledge management, CRM, live chat, email, and phone communications. Without MCP, AI agents can only access one system at a time, requiring multiple custom integrations and complex workflow orchestration.
MCP transforms this fragmented landscape by enabling unified AI agents that can seamlessly orchestrate across all customer service platforms. Instead of building separate integrations for each system, organizations deploy a single AI solution that connects through standardized MCP servers to:
Traditional customer service AI operates with limited context, often requiring agents to manually gather information from multiple sources before resolving issues. MCP enables AI agents to access comprehensive customer context in real-time, dramatically improving first-contact resolution rates and customer satisfaction.
When a customer contacts support, MCP-enabled AI agents can instantly retrieve relevant information across all connected systems. This includes purchase history from e-commerce platforms, previous support interactions from ticketing systems, account status from billing systems, and relevant knowledge base articles.
The AI synthesizes this information to provide personalized, context-aware responses without requiring human agents to perform manual lookups.
Complex customer service scenarios often require actions across multiple systems: updating customer records, creating follow-up tasks, sending notifications, and escalating issues based on predefined criteria.
MCP’s tool capabilities enable AI agents to execute these multi-step workflows automatically while maintaining audit trails and compliance requirements.
For example, when processing a refund request, an MCP-enabled AI agent can:
Discover how Comm100 leverages MCP to connect your entire customer service stack into one seamless AI-powered experience.
Request a demo todayThere are quite a few companies that are currently using the MCP standard. Microsoft has made the most comprehensive commitment to MCP integration. In May 2025, Microsoft released native MCP support in Copilot Studio, offering one-click links to any MCP server, new tool listings, streaming transport, and full tracing and analytics.
GitHub, alongside Microsoft, joined the MCP steering committee at Microsoft Build 2025, contributing a registry service for MCP server discovery and management. The company has also integrated MCP across Azure services, Microsoft 365, and developed an official C# SDK for the .NET ecosystem.
OpenAI officially adopted MCP across their products in March 2025. Sam Altman described the adoption of MCP as a step toward standardizing AI tool connectivity, with integration across the ChatGPT desktop app, OpenAI’s Agents SDK, and the Responses API.
Google DeepMind committed to MCP support when Demis Hassabis, CEO of Google DeepMind, confirmed in April 2025 that MCP support in the upcoming Gemini models and related infrastructure, describing the protocol as “rapidly becoming an open standard for the AI agentic era”.
Block (formerly Square) represents the most extensive enterprise implementation of MCP to date. Block developed Goose, an open source, MCP-compatible AI agent that thousands of Block employees use daily.
At Block, they have developed more than 60 MCP servers, with real impact across engineering, design, security, compliance, customer support, and sales workflows.
Development tools companies like Zed, Replit, Codeium, and Sourcegraph have rapidly adopted MCP to enhance their platforms. And, a bunch of major software companies have also launched their own MCP servers:
Amazon Web Services announced MCP servers for AWS Lambda, Amazon Elastic Container Service (ECS), Amazon Elastic Kubernetes Service (EKS), and Finch, while Cloudflare has become a key infrastructure provider for hosting remote MCP servers at scale.
For companies that are looking to deploy MCP servers, there are a few best practices to consider:
Enterprise MCP deployments require robust security frameworks that go beyond the protocol’s basic specifications. Authentication represents the most critical implementation challenge, as MCP servers often need access to sensitive business data and systems.
Organizations should implement OAuth 2.1 for remote servers and secure credential storage for local implementations, avoiding environment variables or plaintext configuration files.
Access control must be granular, and role-based. Tools that mix both read and write operations make it harder for users to judge the risk accurately. Each tool should stick to a single risk level: read-only (low risk) or non-read (higher risk).
Servers should run behind proper authentication gateways, implement rate limiting, and use encrypted communication channels. Organizations should also consider implementing MCP server allowlists to prevent unauthorized or potentially malicious servers from being used by AI agents.
MCP implementations must account for the token limitations and processing constraints of large language models. LLMs have finite context windows.
For example, Claude Sonnet 4 (the latest version as of writing) can process max 200K input tokens. As tool developers, you are in the best position to know which tool calls produce large outputs and implement checks inside the tool to guard against overflows.
Effective pagination and data filtering become essential for tools that might return large datasets.
Rather than returning complete database dumps or file contents, MCP servers should implement intelligent summarization and provide mechanisms for AI agents to request specific subsets of data when needed.
Caching strategies can significantly improve performance for frequently accessed resources. Local MCP servers should cache static data like documentation or configuration files, while remote servers can implement distributed caching to reduce API calls to underlying systems.
When building custom MCP servers, organizations should start with workflow-driven design rather than exposing raw API endpoints.
It’s usually better to start top-down from the workflow that needs to be automated, and work backwards (in as few steps as possible) to define tools that support that flow effectively. Combine multiple internal API calls into a single high-level tool.
Container-based deployment using Docker provides isolation, consistent environments, and simplified management for MCP servers. This approach addresses security concerns while making it easier to deploy and update servers across different environments.
Organizations should establish clear governance around MCP server development, including code review processes, security assessments, and approval workflows for new servers. This becomes particularly important as MCP adoption scales across different teams and use cases.
The Model Context Protocol represents a fundamental shift in how AI systems connect to the broader technology ecosystem. The evidence is compelling: some of the largest names in the world are already adopting the MCP standard, and if history is any guide, it doesn’t take long before such technologies become the default.
The customer service industry exemplifies MCP’s transformative potential. Instead of maintaining fragmented integrations across ticketing systems, CRM platforms, and communication channels, organizations can deploy unified AI agents that orchestrate seamlessly across their entire technology stack.
This represents the future of business automation: intelligent systems that adapt to existing workflows rather than forcing organizations to rebuild around new tools.
At Comm100, that’s what we do. Our omnichannel AI-powered customer support platform enables organizations like Humana to scale personalized service across all channels.
See how our AI-driven platform adapts to your workflows and scales personalized service.
Request a demo today
By clicking "Subscribe", you agree to our Privacy Policy.
The customer service landscape is experiencing a revolutionary transformation. AI chatbot usage has surged from 62.7% to 73.8% this year alone, and this + Read More
Student counseling services face an unprecedented challenge: according to a 2023 survey from the Healthy Minds Network, 41% of college students reported having + Read More
With over a billion monthly users and 500 million daily active users, Telegram is one of the largest messaging apps in the world. + Read More
