You have probably heard that AI is changing how businesses work. You may have already bought into an AI platform or two. But there is a gap between having access to an AI model and having that model do useful work inside your organisation. That gap is integration, and Model Context Protocol (MCP) is the open standard that closes it.
The problem MCP solves
Before MCP, every AI integration was bespoke. If you wanted your AI platform to read your CRM, you built a custom connector. If you also wanted it to search your knowledge base, that was a separate build. Each platform had its own approach to tool use and data access, which meant integration work was duplicated across providers and fragile to maintain. Tie all that work to one vendor and you have locked yourself in.
MCP standardises this. Instead of building a custom integration for every combination of AI model and business tool, you build one MCP server for each system. That server then works with any AI client that supports the protocol. Because the standard is open, the integrations you build stay yours: they are not tied to a single model or vendor.
What native connectors already handle
An important distinction: modern AI platforms already ship dozens of pre-built connectors. Gmail, Google Drive, Slack, Notion, Jira, GitHub, HubSpot, Stripe, Microsoft 365 (Outlook, SharePoint, OneDrive, Teams), and many more. These are typically one-click OAuth connections any user can enable.
If your systems are covered by off-the-shelf connectors, you do not need custom MCP development for those systems. Your team can connect to Slack in 30 seconds. The value is not in making that connection. It is in configuring which channels the AI can access, setting workspace permissions so departments see only what they should, building prompts that match how your team works, and training people on what to ask.
When you need custom MCP servers
Every mid-market company has systems that will never appear in a connector directory. A 15-year-old SQL Server database running core operations. An industry-specific platform with 500 customers globally. A homegrown CRM or quoting tool. A legacy finance system on SOAP. A niche vertical SaaS product like Zoho CRM or sector-specific platforms.
These systems contain the most valuable, most unique data in your business. Building MCP servers for them requires understanding the system, mapping the data model, handling authentication, and often working with poor or no documentation. This is genuine engineering work. It is also where the biggest productivity gains come from, because this data was previously inaccessible to AI entirely. Build it once against the open protocol and the capability is yours to own, not something you rent.
Who backs MCP
MCP was created by Anthropic and open-sourced in late 2024. Since then, providers and tooling vendors across the industry have adopted the protocol. In early 2025, MCP was transferred to the Agentic AI Foundation under the Linux Foundation, making it a true open standard governed by a neutral body rather than any single vendor.
The adoption numbers are significant. MCP has surpassed 97 million monthly SDK downloads. There are over 10,000 MCP servers available across the ecosystem. This is not a niche experiment. It is becoming the default, vendor-neutral way AI connects to external systems.
What MCP does not do
MCP is a protocol, not a product. It defines how AI and systems communicate, but somebody still needs to build and host the MCP servers for your specific systems. The protocol makes this work standardised, reusable, and portable across models, but it does not eliminate the engineering.
MCP also does not solve adoption. Connecting AI to your systems is necessary but not sufficient. If your team does not know which questions to ask, which workflows AI can improve, or how to structure prompts for their specific work, the integration sits unused. The connector is the plumbing. Adoption is the last mile. A connected pilot is not yet a production system that people rely on every day.
What you should do next
Start by reviewing the native connectors your AI platform already offers. You may find that several of your key systems already have one-click connections available. For those systems, the work is configuration and training, not engineering.
Then map the systems that are not covered: your internal databases, legacy platforms, and bespoke tools. Those are the ones that need custom MCP servers, and they are likely where the biggest value sits. Because MCP is model-agnostic, those servers keep working even if you change or add AI providers later. The organisations getting the most from AI right now are the ones that treated both integration and adoption as first-class problems, not afterthoughts.
Start here
Book a call. We will show you what production AI looks like for your business.
Tell us where you are: stalled pilots, a platform you are evaluating, or teams you want building. We will map the highest-value use cases and show you a realistic path to production. No obligation, no slides.