What Is MCP and Why Does It Matter?
Model Context Protocol (MCP) is an open standard published by Anthropic that defines how AI agents communicate with external tools, data sources, and services. Before MCP, every AI integration required custom glue code — a bespoke adapter for each tool, each API, each data source. MCP replaces that fragmentation with a single, well-defined protocol that any agent and any tool can speak natively.
In practical terms: if your application exposes an MCP server, any MCP-compatible AI agent — Claude, n8n's AI Agent node, or a custom agent you build — can discover your tools, understand their parameters, call them, and handle the results, without any custom integration work on the agent side.
How MCP Works
MCP defines three core primitives:
- Tools — Functions the agent can call. A tool has a name, a description the model uses to decide when to invoke it, and a typed parameter schema. The agent calls the tool; the MCP server executes it and returns the result.
- Resources — Data the agent can read. Files, database records, API responses — anything you expose as a resource is available to the agent as context without requiring a tool call.
- Prompts — Reusable prompt templates the agent can request from the server, enabling server-side prompt management that stays in sync across all agent deployments.
Why MCP Changes the Agent Integration Story
Before MCP, building an agent that could query your database, send a Slack message, and update a CRM record required three separate integrations — each with its own authentication, error handling, and maintenance burden. With MCP, you build one MCP server that exposes all three capabilities as typed tools. Any MCP-compatible agent immediately has access to all three, with no additional integration work.
This is exactly the architecture behind Vanta Embed Agent: you configure your application's APIs as MCP tools via the visual interface, and the embedded agent can call them immediately. The NPM SDK handles the MCP transport layer transparently.
MCP Adoption in 2026
The adoption curve has been steep. Claude Code speaks MCP natively. n8n's AI Agent node supports MCP servers as tool sources. Cursor, Zed, and most major AI coding tools have added MCP support. The ecosystem of community-built MCP servers — for GitHub, PostgreSQL, Slack, Notion, Stripe, and hundreds of other services — means most integrations your agent needs already have a ready-made server you can run or reference.
Building Your First MCP Server
// Minimal MCP server using the official TypeScript SDK
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { z } from "zod";
const server = new McpServer({ name: "my-app", version: "1.0.0" });
server.tool(
"get_order_status",
"Retrieve the current status of an order by ID",
{ orderId: z.string().describe("The order ID to look up") },
async ({ orderId }) => {
const order = await db.orders.findById(orderId);
return { content: [{ type: "text", text: JSON.stringify(order) }] };
}
);
await server.connect(new StdioServerTransport());With this server running, any MCP-compatible agent can call get_order_status with a natural language request like "What is the status of order #1234?" — no additional prompt engineering required.
MCP Security Considerations
Because MCP gives agents real operational reach — the ability to read data and trigger actions — security must be designed in from the start:
- Authentication — MCP servers should verify caller identity before executing tools. Use API keys or OAuth tokens, not open endpoints.
- Authorisation — Each tool should enforce the same access controls your application enforces for human users. The agent is not privileged by default.
- Rate limiting — Prevent runaway agent loops from exhausting resources by rate-limiting tool calls at the MCP server level.
- Audit logging — Log every tool invocation with the calling agent's identity, parameters, and result. This is your compliance audit trail.
Frequently Asked Questions
Is MCP only for Claude models?
No. MCP is an open protocol. While Anthropic published the specification, any model or agent framework can implement it. n8n, OpenHands, Cursor, and multiple open-source agent frameworks have native MCP support.
How does MCP differ from OpenAI's function calling?
OpenAI's function calling is model-specific and stateless — tools are defined per-request in the API payload. MCP is a persistent server protocol: tools are defined once, discoverable by any compatible agent, and can maintain state across calls. MCP also supports resources and prompts, which function calling does not.
Do I need to run an MCP server to use Claude Code?
No — Claude Code works without any MCP server. But if you want to give Claude Code access to custom tools specific to your application (querying your internal APIs, reading from proprietary data sources), an MCP server is the right way to expose those capabilities.
Conclusion
MCP is the missing interoperability layer that makes agent systems practical to build and maintain at scale. It replaces custom integration code with a well-defined protocol, makes agent capabilities discoverable rather than hardcoded, and gives every tool you build instant compatibility with the growing ecosystem of MCP-native agents. At Vantaverse, MCP is now a standard component of every agent system we ship.