Remote MCP Server Overview

This page introduces Remote MCP servers and helps you decide if they’re right for your use case.

After reading this page, you will be able to:

Explain what a Remote MCP server is and how tools differ from pipelines
Identify use cases where Remote MCP provides business value
Describe how MCP tools expose Redpanda Connect components to AI

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI agents use tools. Think of it like a universal adapter: instead of building custom integrations for every AI system, you define your tools once using MCP, and any MCP-compatible AI client can discover and use them.

Without MCP, connecting AI to your business systems requires custom API code, authentication handling, and response formatting for each AI platform. With MCP, you describe what a tool does and what inputs it needs, and the protocol handles the rest.

What is Remote MCP?

Remote MCP lets you build and host MCP servers in your Redpanda Cloud clusters. Your tools run next to your data, managed by Redpanda, so you get:

Always-on availability: No local process to run. Your tools are hosted and managed by Redpanda Cloud.
Proximity to data: Tools execute next to your cluster for lower latency and simpler networking.
Secure secrets management: Use the Secrets Store instead of hardcoding credentials.
Fast iteration: Define tools as YAML, deploy, and your AI agents can use them immediately.

MCP tools are not pipelines

If you already use Redpanda Connect, you might wonder how MCP tools differ from pipelines.

A pipeline is a continuous data flow: data streams from an input, through processors, to an output. The pipeline runs indefinitely, processing many messages over time.

An MCP tool is different. It’s a single component that executes on demand when called by an AI client. The tool starts, runs, and completes for each invocation. There is no persistent state between calls. Think of it like calling a function rather than running a service.

This request/response pattern is what makes MCP tools useful for AI agents: the agent asks a question, the tool runs, and it returns an answer.

Use cases

Category	Example prompts
Operational monitoring	Check partition lag for customer-events topic Show me the top 10 producers by message volume today Get schema registry health status
Data enrichment and analysis	Fetch user profile data and recent orders for customer ID 12345 Get real-time stock prices for symbols in my portfolio topic Analyze sentiment of latest product reviews
Team productivity	Deploy my microservice to the staging environment Generate load test data for the payments service Create a summary dashboard of this week’s incident reports
Business intelligence	What are the trending products in the last 24 hours? Show revenue impact of the latest feature deployment Get customer satisfaction scores from support tickets

How it works

Remote MCP servers sit between AI clients and your data:

Your AI agent connects to your MCP server using rpk cloud mcp proxy or direct authentication
A user asks their AI agent something like "What’s the weather in London?"
The server finds the matching tool and runs your Redpanda Connect configuration
Your configuration fetches data, transforms it, and returns a structured response
The AI agent gets the data and can use it to answer the user

What a tool looks like

A tool is a YAML configuration with two parts: the logic (what the tool does) and the metadata (how AI understands it).

Here’s a minimal example that returns weather data:

http:
  url: "https://wttr.in/${! this.city }?format=j1"
  verb: GET

meta:
  mcp:
    enabled: true
    name: get_weather
    description: "Get current weather for a city"
    properties:
      - name: city
        type: string
        description: "City name"
        required: true

When an AI client asks about weather, it calls this tool with the city name. The tool fetches data from the weather API and returns it.

MCP specification support

MCP servers implement the open MCP protocol for tool exposure. Only the tool concept from the MCP server specification is supported. Features such as MCP resources and prompts are not yet available.

For full details, see the official MCP server specification.