Use Cases

MCP Servers

MCP (Model Context Protocol) servers enable AI assistants and LLMs to access external tools, data sources, and capabilities in a standardized way. They act as bridges between AI models and real-world resources—file systems, databases, APIs, search engines, and custom business logic. As AI assistants become more capable, MCP servers are becoming critical infrastructure for grounding AI in reality.

With Unikraft Cloud, you are enhancing your MCP servers with instant deployment, scale-to-zero economics, and millisecond-level cold starts—perfect for the on-demand nature of AI interactions.

Why run MCP servers on Unikraft Cloud

⚡ Zero-latency context injection

AI assistants need instant access to tools and data. Every millisecond of latency impacts the user experience. Unikraft Cloud's unikernel architecture reduces OS overhead:

Single-digit-millisecond cold starts mean MCP servers are ready instantly.
Minimal memory footprint allows dense packing of specialized servers.

🔒 Isolated execution

MCP servers often access sensitive data—file systems, databases, proprietary APIs. Unikraft Cloud provides VM-level isolation for each server instance:

Strong isolation prevents cross-contamination between different MCP contexts.
Immutable unikernel images mitigate runtime tampering risks.
Minimal attack surface with only essential components included.

💸 Pay-per-query economics

AI interactions are inherently bursty—periods of intense activity followed by silence. Traditional server architectures waste resources during idle periods. With Unikraft Cloud:

MCP servers scale-to-zero when not in use, costing nothing.
Instant wake-up ensures your queries have no noticeable delay.
Fine-grained billing means you only pay for actual AI interactions.

🌍 Edge-ready deployment

Unikraft Cloud help you reduce the latency of your applications:

Choose to deploy to your closest metro.
Lightweight footprint enables cost-effective multi-region deployment.
Fast startup times ensure consistent performance across all locations.

Getting started

MCP servers on Unikraft Cloud provide the perfect runtime for AI tool integration, being fast, secure, and cost-efficient. Scale-to-zero economics meet instant availability, ensuring your AI assistants always have the tools they need, exactly when they need them.

This guide shows you how to deploy two types of MCP servers:

Example 1: ArXiv MCP server

The arXiv MCP Server is a community-made MCP server that connects AI tools directly to the arXiv archive of scientific articles and publications. It allows MCP clients to browse and analyze papers.

To run an example based on it, follow these steps:

Install the kraft CLI tool and a container runtime engine (for example, Docker).
Clone the examples repository and cd into the examples/mcp-server-arxiv/ directory:

Code
 
git clone https://github.com/unikraft-cloud/examples
cd examples/mcp-server-arxiv/

Make sure to log into Unikraft Cloud by setting your token and a metro close to you. This guide uses fra (Frankfurt, 🇩🇪):

Code
 
# Set Unikraft Cloud access token
export UKC_TOKEN=token
# Set metro to Frankfurt, DE
export UKC_METRO=fra

Follow the instructions in the example's README to set it up and grab the server's FQDN.

You can use the included example client to test the server, or connect to it using your favourite MCP client. For example, using Visual Studio Code, you can add the following to your .vscode/mcp.json file:


Code
 
{
    "servers": {
        "arxiv-mcp": {
            "type": "http",
            "url": "https://<FQDN>/mcp"
        }
    }
}

The tool will then show up in Copilot Chat. You can find similar instructions for Claude Code and other MCP-compatible clients.

Example query that will trigger a tool call:

Q: Could you download the original Unikraft paper for me?

A: The original Unikraft paper (“Unikraft: Fast, Specialized Unikernels the Easy Way”) is being downloaded and converted. I will notify you as soon as it is ready for reading or further actions.

Q: Dump the whole first section of the paper here.

A: Here is the full first section (“Introduction”) from the Unikraft paper:

1 Introduction

Specialization is arguably the most effective way to achieve outstanding performance, whether it is for achieving high throughput in network-bound applications, making language runtime environments more efficient, or providing efficient container environments, to give some examples. Even in the hardware domain, and especially with the demise of Moore’s law, manufacturers are increasingly leaning towards hardware specialization to achieve ever better performance; the machine learning field is a primary exponent of this. ...

Example 2: Custom MCP server

This example demonstrates how to build a minimal custom MCP server from scratch. It provides weather information and time utilities. This shows the core patterns for building your own MCP tools.

To run it, follow these steps:

Install the kraft CLI tool and a container runtime engine (for example, Docker).
Clone the examples repository and cd into the examples/mcp-server-simple/ directory:

Code
 
git clone https://github.com/unikraft-cloud/examples
cd examples/mcp-server-simple/

Make sure to set your Unikraft Cloud token and metro:

Code
 
export UKC_TOKEN=token
export UKC_METRO=fra

Follow the instructions in the example's README to set it up and grab the server's FQDN.


Code
 
{
    "servers": {
        "custom-mcp": {
            "type": "http",
            "url": "https://<FQDN>/mcp"
        }
    }
}

The tool will then show up in Copilot Chat. You can find similar instructions for Claude Code and other MCP-compatible clients.

Example query that will trigger a tool call:

Q: What is the current weather in Paris, and what time is it there?

A: The current weather in Paris is 21°C, partly cloudy, with 62% humidity. The local time in Paris is 23:21 (11:21 PM), Tuesday, December 2, 2025 (UTC+1).

Building your own MCP server

MCP servers follow a simple pattern:

Define tools: Specify what capabilities your server provides (functions, APIs, data sources).
Handle requests: Process MCP protocol requests from AI assistants.
Return results: Format responses in the MCP protocol format.

Common use cases for custom MCP servers:

Database access: Query internal databases, data warehouses, or analytics platforms.
API integration: Connect AI assistants to third-party services (customer relationship management, ticketing, monitoring).
Custom business logic: Integrate company-specific calculations, validations, or workflows.
Document retrieval: Search and retrieve from internal knowledge bases or document stores.

Unikraft Cloud's instant deployment and scale-to-zero model makes it ideal for hosting dozens of specialized MCP servers.

Learn more

Use the --help option for detailed information on using Unikraft Cloud:

Code
 
kraft cloud --help

Or visit the CLI Reference.

For more information about the Model Context Protocol:

Edit this page

Last modified on February 4, 2026

Headless Browsers Serverless Databases

Use Cases

MCP Servers

With Unikraft Cloud, you are enhancing your MCP servers with instant deployment, scale-to-zero economics, and millisecond-level cold starts—perfect for the on-demand nature of AI interactions.

Why run MCP servers on Unikraft Cloud

⚡ Zero-latency context injection

AI assistants need instant access to tools and data. Every millisecond of latency impacts the user experience. Unikraft Cloud's unikernel architecture reduces OS overhead:

Single-digit-millisecond cold starts mean MCP servers are ready instantly.
Minimal memory footprint allows dense packing of specialized servers.

🔒 Isolated execution

MCP servers often access sensitive data—file systems, databases, proprietary APIs. Unikraft Cloud provides VM-level isolation for each server instance:

Strong isolation prevents cross-contamination between different MCP contexts.
Immutable unikernel images mitigate runtime tampering risks.
Minimal attack surface with only essential components included.

💸 Pay-per-query economics

AI interactions are inherently bursty—periods of intense activity followed by silence. Traditional server architectures waste resources during idle periods. With Unikraft Cloud:

MCP servers scale-to-zero when not in use, costing nothing.
Instant wake-up ensures your queries have no noticeable delay.
Fine-grained billing means you only pay for actual AI interactions.

🌍 Edge-ready deployment

Unikraft Cloud help you reduce the latency of your applications:

Choose to deploy to your closest metro.
Lightweight footprint enables cost-effective multi-region deployment.
Fast startup times ensure consistent performance across all locations.

Getting started

This guide shows you how to deploy two types of MCP servers:

Example 1: ArXiv MCP server

The arXiv MCP Server is a community-made MCP server that connects AI tools directly to the arXiv archive of scientific articles and publications. It allows MCP clients to browse and analyze papers.

To run an example based on it, follow these steps:

Install the kraft CLI tool and a container runtime engine (for example, Docker).
Clone the examples repository and cd into the examples/mcp-server-arxiv/ directory:

Code
 
git clone https://github.com/unikraft-cloud/examples
cd examples/mcp-server-arxiv/

Make sure to log into Unikraft Cloud by setting your token and a metro close to you. This guide uses fra (Frankfurt, 🇩🇪):

Code
 
# Set Unikraft Cloud access token
export UKC_TOKEN=token
# Set metro to Frankfurt, DE
export UKC_METRO=fra

Follow the instructions in the example's README to set it up and grab the server's FQDN.


Code
 
{
    "servers": {
        "arxiv-mcp": {
            "type": "http",
            "url": "https://<FQDN>/mcp"
        }
    }
}

The tool will then show up in Copilot Chat. You can find similar instructions for Claude Code and other MCP-compatible clients.

Example query that will trigger a tool call:

Q: Could you download the original Unikraft paper for me?

A: The original Unikraft paper (“Unikraft: Fast, Specialized Unikernels the Easy Way”) is being downloaded and converted. I will notify you as soon as it is ready for reading or further actions.

Q: Dump the whole first section of the paper here.

A: Here is the full first section (“Introduction”) from the Unikraft paper:

1 Introduction

Specialization is arguably the most effective way to achieve outstanding performance, whether it is for achieving high throughput in network-bound applications, making language runtime environments more efficient, or providing efficient container environments, to give some examples. Even in the hardware domain, and especially with the demise of Moore’s law, manufacturers are increasingly leaning towards hardware specialization to achieve ever better performance; the machine learning field is a primary exponent of this. ...

Example 2: Custom MCP server

This example demonstrates how to build a minimal custom MCP server from scratch. It provides weather information and time utilities. This shows the core patterns for building your own MCP tools.

To run it, follow these steps:

Install the kraft CLI tool and a container runtime engine (for example, Docker).
Clone the examples repository and cd into the examples/mcp-server-simple/ directory:

Code
 
git clone https://github.com/unikraft-cloud/examples
cd examples/mcp-server-simple/

Make sure to set your Unikraft Cloud token and metro:

Code
 
export UKC_TOKEN=token
export UKC_METRO=fra

Follow the instructions in the example's README to set it up and grab the server's FQDN.


Code
 
{
    "servers": {
        "custom-mcp": {
            "type": "http",
            "url": "https://<FQDN>/mcp"
        }
    }
}

The tool will then show up in Copilot Chat. You can find similar instructions for Claude Code and other MCP-compatible clients.

Example query that will trigger a tool call:

Q: What is the current weather in Paris, and what time is it there?

A: The current weather in Paris is 21°C, partly cloudy, with 62% humidity. The local time in Paris is 23:21 (11:21 PM), Tuesday, December 2, 2025 (UTC+1).

Building your own MCP server

MCP servers follow a simple pattern:

Define tools: Specify what capabilities your server provides (functions, APIs, data sources).
Handle requests: Process MCP protocol requests from AI assistants.
Return results: Format responses in the MCP protocol format.

Common use cases for custom MCP servers:

Database access: Query internal databases, data warehouses, or analytics platforms.
API integration: Connect AI assistants to third-party services (customer relationship management, ticketing, monitoring).
Custom business logic: Integrate company-specific calculations, validations, or workflows.
Document retrieval: Search and retrieve from internal knowledge bases or document stores.

Unikraft Cloud's instant deployment and scale-to-zero model makes it ideal for hosting dozens of specialized MCP servers.

Learn more

Use the --help option for detailed information on using Unikraft Cloud:

Code
 
kraft cloud --help

Or visit the CLI Reference.

For more information about the Model Context Protocol:

Edit this page

Last modified on February 4, 2026

Headless Browsers Serverless Databases