Model Context Protocol (MCP)

MCP as the universal adapter that lets AI models plug into any data source. Released by Anthropic.
MCP doesn't focus on agents talking to agents—instead, it standardizes how AI models connect to tools and data.
MCP provides a universal, open standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol. It's like giving your AI assistant a master key to every database, API, and tool in your organization.

In short, we can say that it is the standardized version of connecting LLMs to external data sources.

This visual was inspired by documentation from Anthropic and was created by the Black Kite Data Research team.

How It Works

MCP works on a client ↔ server handshake: the server advertises capabilities (tools, resources, prompts) through JSON-RPC, and the client forwards those to the LLM host. Your class is one such MCP Server that surfaces a database. (The "server" concept here does not have to be a real HTTP service; it can also be a local subprocess.).

Code element	What MCP expects	Narrative for Readers
self.name (e.g., "company_database")	A unique server ID the client can display in its capability list.	Think of name as the USB descriptor your agent shows when it’s plugged in.
self.resources list	Static registry of data blobs (not functions). Each item will later map to a resource object in the MCP spec.	These are the tables you’re willing to expose.
Async handle_resource_request()	Implements the Resources RPC; the client calls it when the LLM wants rows.	One coroutine per resource keeps the server non-blocking.
describe_capabilities()	Mandatory initialize() response: advertises resources (or tools) with JSON-Schema so the LLM can validate its own requests.	Without this, the model has no contract—schemas are your guard-rails.

Use Case

The Challenge: When a Major Cloud Provider Gets Compromised

Imagine it's 3 AM, and news breaks that a major cloud infrastructure provider has suffered a data breach. Your TPRM team needs to immediately answer: "Which of our 500+ vendors are affected, what's their exposure level, and what's our business impact?"

Traditional Approach Problems:

6 different vendor databases with separate APIs
Security ratings from 3 different providers (each with custom integration)
Financial data from D&B, compliance records from audit systems
Manual correlation taking 4-6 hours while business risk escalatesHey, you just created a text paragraph! Somebody once said that the pen is mightier than the sword — and that was in 1839. Just imagine, with the power of cutting-edge content experiences and the ability to distribute your content around the world in mere seconds, writing this paragraph could be one of the most influential things you ever do!

The MCP-Enabled Solution: Instant Vendor Impact Intelligence

One simple query instead of 6 different API calls !

MCP is an ideal choice in this scenario because it requires standardized data access from multiple sources, not agent-to-agent coordination (A2A) or complex workflow orchestration (LCEL). When a breach occurs, you need instant data collection from vendor databases, security ratings, and compliance systems; MCP’s “universal connector” approach combines 6+ custom API integrations into a single standardized interface that can pull real-time data from all sources simultaneously.