Anthropic MCP Server

In November Anthropic released a standard called MCP servers. I didnt realise then how valuable these servers are for Agentic AI, this is what they are and why they are valuable for enterprise:

𝗪𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗮𝗿𝗲:

MCP servers are an API gateway that sit on top of of your business APIs, data sources (structured and unstructured) and prompts. Inside the MCP server you can construct tools that LLMs or Agents can use to assist with their tasks. The server acts as a 'man in the middle' between your agent and LLMs and the data and tools it needs.

For instance, if you wanted an agent to be able to get current orders and create new orders for a customer, then these could be two MCP tools 'create_order(customer_id)' and 'list_orders(customer_id)' which make specific SQL requests to your databases & have suitable validation in place.

Anthropic have released many different implementations of their MCP standard, which are available on the repo:

https://github.com/modelcontextprotocol

What I see as the benefits of adopting a framework like this are:

𝗦𝗶𝗻𝗴𝗹𝗲 𝗣𝗼𝗶𝗻𝘁 𝗼𝗳 𝗔𝗰𝗰𝗲𝘀𝘀

By having a single point of access it makes any resources created for LLMs useable by multiple different agents & all resources are visible in a single place with Anthropics desktop client.

𝗔𝗯𝘀𝘁𝗿𝗮𝗰𝘁 𝗰𝗼𝗺𝗽𝗹𝗲𝘅 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀 𝗹𝗼𝗴𝗶𝗰

Business APIs are often very messy. By conciously constructing an API Gateway with Agents and LLMs in mind you can connect and tailor agents ability to interact with your business logic.

𝗙𝗮𝘀𝘁𝗲𝗿 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁

If all of your resources and tools are created to be generally accessible, it means that you can create ‘off the shelf’ agents using tools like AWS bedrock and Azure AI Studio to quickly create agents as and when they are required.

𝗨𝘀𝗲𝗿 𝘀𝗽𝗲𝗰𝗶𝗳𝗶𝗰 𝗔𝗰𝗰𝗲𝘀𝘀

You can limit access to tools and resources depending on the user id passed in so that your agent doesnt have to worry about permissions. This also means that you can store memory about previous user actions in one place for all agents to access.

‍