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
| Component | What it does | Example |
|---|---|---|
| MCP Host | The AI application that needs to access tools | Claude Code, Cursor, a custom AI assistant |
| MCP Client | Maintains the connection between host and server | Built into the AI application |
| MCP Server | Exposes tools and data through the standard protocol | Umbraco MCP server, GitHub MCP server, database MCP server |
| Resources | Data the AI can read (like GET requests) | CMS content, database records, file contents |
| Tools | Actions the AI can perform (like POST requests) | Create a page, update a record, send a notification |
Real-world MCP use cases
| Use case | What happens | Business benefit |
|---|---|---|
| CMS management | AI reads, creates, and updates website content directly in Umbraco | Content updates in minutes instead of hours |
| Code assistance | AI accesses your codebase, runs tests, and deploys changes | Faster development cycles, fewer context switches |
| Data analysis | AI queries your database and generates reports conversationally | No SQL knowledge needed, instant insights |
| Project management | AI reads tickets, updates status, and creates tasks | Less admin, more focus on actual work |
| Customer support | AI accesses your knowledge base and CRM to answer customer queries | Faster, more accurate responses grounded in your data |
MCP vs traditional API integrations
| Factor | Traditional API integration | MCP |
|---|---|---|
| Setup effort | Custom code per tool per AI | One standard protocol, plug and play |
| Maintenance | Each integration maintained separately | Update the MCP server, all AI clients benefit |
| AI provider lock-in | Built for one specific AI | Works with any MCP-compatible AI |
| Discovery | Developer must know what APIs exist | AI can discover available tools automatically |
| Security | Varies by implementation | Built-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.
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
| Component | What it does | Example |
|---|---|---|
| MCP Host | The AI application that needs to access tools | Claude Code, Cursor, a custom AI assistant |
| MCP Client | Maintains the connection between host and server | Built into the AI application |
| MCP Server | Exposes tools and data through the standard protocol | Umbraco MCP server, GitHub MCP server, database MCP server |
| Resources | Data the AI can read (like GET requests) | CMS content, database records, file contents |
| Tools | Actions the AI can perform (like POST requests) | Create a page, update a record, send a notification |
Real-world MCP use cases
| Use case | What happens | Business benefit |
|---|---|---|
| CMS management | AI reads, creates, and updates website content directly in Umbraco | Content updates in minutes instead of hours |
| Code assistance | AI accesses your codebase, runs tests, and deploys changes | Faster development cycles, fewer context switches |
| Data analysis | AI queries your database and generates reports conversationally | No SQL knowledge needed, instant insights |
| Project management | AI reads tickets, updates status, and creates tasks | Less admin, more focus on actual work |
| Customer support | AI accesses your knowledge base and CRM to answer customer queries | Faster, more accurate responses grounded in your data |
MCP vs traditional API integrations
| Factor | Traditional API integration | MCP |
|---|---|---|
| Setup effort | Custom code per tool per AI | One standard protocol, plug and play |
| Maintenance | Each integration maintained separately | Update the MCP server, all AI clients benefit |
| AI provider lock-in | Built for one specific AI | Works with any MCP-compatible AI |
| Discovery | Developer must know what APIs exist | AI can discover available tools automatically |
| Security | Varies by implementation | Built-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.