Glossary / Software Development

Model Context Protocol (MCP)

An open standard that lets AI assistants connect directly to external tools, databases, and services, turning them from chatbots into intelligent agents.

Definition

Model Context Protocol (MCP) is an open standard created by Anthropic in late 2024 that defines how AI assistants communicate with external tools and data sources. Think of it as a universal adapter: instead of building a custom integration for every tool an AI needs to access, MCP provides a standard way for any AI to talk to any tool. This means an AI assistant can read your CMS content, query your database, check your analytics, or manage your project board, all within a single conversation, without anyone writing bespoke glue code for each connection.

Why MCP matters for businesses

Before MCP, connecting an AI assistant to your business tools required custom API integrations for every single tool. Each integration was expensive to build, fragile to maintain, and locked to one specific AI provider. MCP changes this by creating a shared protocol that any AI assistant can use to connect to any MCP-enabled tool.

For UK businesses, this means you can give AI assistants real access to your systems, not just general knowledge, but your actual data, your actual CMS, your actual customer records. The AI stops being a fancy search engine and starts being a genuinely useful team member.

How MCP works

ComponentWhat it doesExample
MCP HostThe AI application that needs to access toolsClaude Code, Cursor, a custom AI assistant
MCP ClientMaintains the connection between host and serverBuilt into the AI application
MCP ServerExposes tools and data through the standard protocolUmbraco MCP server, GitHub MCP server, database MCP server
ResourcesData the AI can read (like GET requests)CMS content, database records, file contents
ToolsActions the AI can perform (like POST requests)Create a page, update a record, send a notification

Real-world MCP use cases

Use caseWhat happensBusiness benefit
CMS managementAI reads, creates, and updates website content directly in UmbracoContent updates in minutes instead of hours
Code assistanceAI accesses your codebase, runs tests, and deploys changesFaster development cycles, fewer context switches
Data analysisAI queries your database and generates reports conversationallyNo SQL knowledge needed, instant insights
Project managementAI reads tickets, updates status, and creates tasksLess admin, more focus on actual work
Customer supportAI accesses your knowledge base and CRM to answer customer queriesFaster, more accurate responses grounded in your data

MCP vs traditional API integrations

FactorTraditional API integrationMCP
Setup effortCustom code per tool per AIOne standard protocol, plug and play
MaintenanceEach integration maintained separatelyUpdate the MCP server, all AI clients benefit
AI provider lock-inBuilt for one specific AIWorks with any MCP-compatible AI
DiscoveryDeveloper must know what APIs existAI can discover available tools automatically
SecurityVaries by implementationBuilt-in permission model, human approval for actions
Cost to build£2,000 to £10,000 per integration£500 to £3,000 for an MCP server (reusable across all AI tools)

When NOT to use MCP

  • When you do not use AI tools yet: MCP connects AI assistants to your systems. If you are not using AI assistants in your workflow, MCP adds complexity without benefit
  • When a simple API call will do: If you just need one system to send data to another on a schedule, a traditional API integration or webhook is simpler and more reliable
  • When security requirements prohibit AI access: MCP gives AI assistants real access to real systems. If your compliance requirements do not allow AI tools to access certain data, do not connect them via MCP
  • When your tools have no MCP server: MCP only works if the tool you want to connect has an MCP server. The ecosystem is growing fast, but not every tool has one yet

How we use MCP at Original Objective

We built an MCP server for Umbraco CMS that lets AI assistants manage content, create pages, update properties, and publish changes directly through natural language. This is not a demo. It is how we actually build and maintain our own website. If you want to see how MCP could connect your AI tools to your business systems, book a free 30-minute discovery call.

Frequently Asked Questions

Common questions about Model Context Protocol and AI tool integration.

Is MCP only for developers?

Setting up MCP servers requires technical knowledge, but using them does not. Once an MCP server is configured, anyone can interact with the connected tools through natural language. For example, a marketing manager could ask an AI assistant to "update the homepage headline" and the AI would make the change directly in the CMS via MCP, no coding required.

Is MCP secure? Can the AI do anything it wants?

MCP has a built-in permission model. Each MCP server defines what tools and resources are available, and most implementations require human approval before the AI takes destructive actions like deleting content or modifying data. You control what the AI can access and what actions require your confirmation. It is no different from giving a team member access to specific systems with specific permissions.

Does MCP work with all AI assistants?

MCP is an open standard, so any AI assistant can implement it. Currently, the strongest support is in Anthropic's Claude (including Claude Code and the Claude desktop app), Cursor, and other developer tools. OpenAI has also announced MCP support for ChatGPT. As the standard matures, expect most major AI tools to support it. The beauty of the standard is that an MCP server you build today will work with any future AI assistant that supports the protocol.