By Doug Silkstone — 17 Mar 2025

Why Your AI Is Trapped in a Box (And How to Set It Free)

Over the last two years, the hype cycle in practical AI has been wild - from "this is completely useless" to "this will replace all developers" and back again. The reality is far more interesting, and it depends entirely on whether your AI can actually do anything beyond generating text.

What's become clear to me that AI is genuinely useful when interacting with our tools and data. It needs to live outside the box. In the real world space. That's where MCP comes in, eventually, and this is where I'm quietly, confidently, leading the way with my thoughts.

TL;DR: Here's the template. Go get building! https://github.com/dougwithseismic/withseismic-mcp

Nobody's saying it out loud, but I will: AI without access to real-world tools is just an expensive autocomplete.

We've watched AI evolve from generating basic images to writing entire articles. But while everyone's been obsessed with prompt engineering, the real revolution has been quietly building: the Human<>API layer that connects AI systems to both digital tools and human expertise.

The global API marketplace is valued at $18 billion in 2024 and projected to reach nearly $50 billion by 2030. Within this market, the AI API segment is growing at a staggering 32.2% CAGR, expected to hit $179 billion by 2030. This isn't just another tech bubble – it's the infrastructure for fundamentally reshaping how we work.

What Is MCP And Why It's An Actual "Game-Changer"

Model Context Protocol (MCP) lets AI models use external tools and data. In simple terms, it gives AI the ability to break free from its text box and actually do things in the real world.

Why does it this matter? Without MCP, AI is trapped in an isolation chamber. It can discuss the JIRA tickets you show it, but can't create one. It can talk about your calendar but can't schedule a meeting. It can suggest code changes but can't commit them. It doesn't have context.

MCP creates the connection layer that changes this. Today, this means tools like Cursor and Claude Desktop can:

Access your company databases in real-time
Call your internal APIs with proper authenticationTrigger automated workflows in your existing tools
Create, update, and delete actual business information
Coordinate human tasks when human expertise is needed

Not a bad start, but we still have a lot to do. First off, this all needs to be built by someone. Whether that someone is a human or an LLM is where things get interesting—and this is where I'm focused.

Automating the Bureaucratic Nightmare

Let's talk about what this really means. The average knowledge worker spends 2.5 hours daily on administrative busywork. Forms, timesheets, expense reports, compliance documentation – the modern bureaucratic nightmare. If this isn't you, then consider yourself lucky.

With MCP-enabled systems, this changes dramatically:

Bureaucratic Forms Processed in Seconds

Imagine your AI assistant handling every government form, insurance claim, or business registration automatically:

It extracts information from your existing documents
It fills out required forms with perfect accuracy
It validates against submission requirements
It submits through official APIs with proper authentication
It follows up on status and alerts you only when human input is needed

No more spending hours deciphering confusing requirements or re-entering the same information on multiple forms. An MCP-enabled assistant with the right tools can reduce a two-hour paperwork session to a 30-second conversation.

Companies like Conrad Electronic are already using similar API-enabled tools that 60% of their customers have adopted, dramatically streamlining operations.

The New Labor Market: AI Bidding for Human Time

Perhaps the most revolutionary aspect of the Human<>API layer is how it transforms labor markets. We're moving toward a world where:

AI systems can autonomously request human assistance
- An AI processing mortgage applications might bid for a human underwriter's time for complex cases
- An AI managing inventory can dispatch human workers to verify physical counts when discrepancies appear
- An AI monitoring security footage can instantly alert human security when unusual patterns emerge
Humans can delegate physical tasks to coordinated AI-human networks
- Need groceries? Your request goes to an AI that coordinates with local delivery networks
- Home repair issue? Your AI diagnoses the problem and dispatches the right specialist.
- Legal document needs review? AI pre-processes everything and connects you with the right expert for final approval

This isn't science fiction. In healthcare, Human API's Clinical API already connects to 85% of US hospitals, covering 264 million patients. The technology supports 18 types of data elements, creating digital experiences that let consumers securely share their medical data with companies and service providers.

Transforming Local Commerce Into Global Networks

The corner store isn't going extinct – it's being transformed into a node in a global fulfillment network. With MCP-enabled systems, every local business becomes both physical and digital:

The neighborhood grocery store becomes an e-commerce endpoint with AI-managed inventory and dynamic pricing
The local carpenter joins a talent network where AI matches their skills to nearby projects
The corner pharmacy integrates with healthcare APIs for automated prescription management

This transformation is already happening. Woolworths launched Everyday Market in Australia – an online marketplace offering products from third-party vendors – and saw 20% growth in new eCommerce shoppers and steady revenue increases, with grocery-adjacent categories outperforming forecasts.

The New Specialist Workforce

As LLMs displace traditional knowledge work, a new category of specialist human tasks emerges – ones that combine human judgment, physical presence, and specialized expertise:

Physical validation specialists who verify AI analyses in the real world
Exception handlers who step in when AI encounters unusual scenarios
Experience designers who craft human-AI interaction frameworks
Trust verification experts who ensure AI systems remain honest and accurate
Last-mile problem solvers who handle the final, most complex 5% of tasks

The growth of these roles is creating entirely new job categories. Rather than mass unemployment, we're seeing a redistribution of labor toward tasks that combine human judgment with AI efficiency.

WithSeismic MCP: The Next.js for Model Context Protocol

The problem until now has been that building MCP servers required significant infrastructure work. That's where WithSeismic MCP Boilerplate comes in – it's like Next.js for MCP, abstracting away the complexity so you can focus on building tools, not infrastructure.

Here's how simple it is to create a tool with WithSeismic MCP:

export const myTool = new Tool({
  name: "myTool",
  description: "What my tool does",
  inputSchema: z.object({ query: z.string() }),
  outputSchema: z.object({ result: z.string() }),
}, async (args) => {
  // Your logic here
  return { result: "Done!" };
});

The system automatically handles:

Tool registration
Input validation
Connection management
Error handling
Logging

This means you spend time building solutions that connect your AI to real business systems instead of wrestling with MCP infrastructure.

What This Enables Right Now

Here's what companies are already building with MCP-based systems:

Knowledge Worker Augmentation

AI that joins your team meetings, transcribes everything, and automatically creates and assigns tasks in your project management system
Document processing that extracts data, validates it against business rules, and pushes it to the right systems
Email triage that not only categorizes messages but takes action on routine requests

Physical World Integration

Retail systems that coordinate between online shoppers and in-store staff for seamless pickup experiences
Healthcare networks where AI processes routine tests and only escalates anomalies to physicians
Manufacturing operations where AI analyzes quality control images and dispatches technicians when needed

New Business Models

On-demand expertise networks where AI handles 80% of customer inquiries and routes the rest to human specialists
Hybrid service offerings where AI handles routine work 24/7 while humans provide premium, personalized service
Marketplace platforms where AI matches supply and demand across physical and digital providers

The Real Question: Why Are You Still Trapped In A Text Box?

The most important statistic: 68% of digital transformation initiatives fail because they focus on the wrong things. Companies obsess over the latest AI models while ignoring the critical integration layer that connects AI to actual business operations.

The question isn't whether AI will transform your business – it's whether you'll be part of the transformation or left behind. MCP isn't just another technical standard; it's the bridge between AI's theoretical potential and practical business impact.

While others are still prompting chatbots for marginally better text, forward-thinking teams are building systems where AI and humans work together seamlessly, each handling the parts they do best. The future isn't AI replacing humans – it's integrated systems that combine AI's scalability with human judgment and expertise.

The tools are here. The market is growing exponentially. The only question is whether you'll continue treating AI as an isolated text generator or start building the connections that transform it into a business force multiplier.

Get started with the WithSeismic MCP template today and stop leaving your AI trapped in a box.

Doug Silkstone is the founder of WithSeismic. Find more at WithSeismic.com.

What is MCP and Why Should You Care?

Model Context Protocol (MCP) lets AI models use external tools and data. Think of it as giving AI the ability to reach beyond its text box and actually do things in the real world.

Here's why that matters: AI models are great at understanding language and generating content, but they're isolated from your actual work environment. They can't natively access your databases, call your APIs, or interact with your team's tools.

MCP solves this problem by creating a connection layer. When your AI needs to check JIRA tickets, send a Slack message, or query your product database, it can do that through MCP.

The excitement around MCP comes from its practical applications:

Engineers can build tools that AI can call directly
Teams can connect their existing systems to AI workflows
The same tools work across different AI interfaces (Claude, ChatGPT, Cursor)
Custom workflows become possible without rebuilding everything

In practice, an MCP is a server that knows about your tools and makes them available to AI. When the AI needs something, it asks the MCP server to run the appropriate tool and returns the results.

The Next.js for Model Context Protocol: WithSeismic MCP

The problem is that building an MCP server typically requires a lot of infrastructure code. That's where the WithSeismic MCP Boilerplate comes in—it's like Next.js for MCP, abstracting away the complexity so you can focus on building tools, not infrastructure.

I'm using it every day with my clients, and I want to share it so you can spin up your own server and get your team working on stuff that actually matters.

What Makes This MCP Implementation Different

The existing examples for MCP are solid starting points, but they often leave you dealing with a lot of infrastructure code. I wanted something where I could just define tools and get back to building.

Here's all you need to create a tool with the WithSeismic MCP:

export const myTool = new Tool({
  name: "myTool",
  description: "What my tool does",
  inputSchema: z.object({ query: z.string() }),
  outputSchema: z.object({ result: z.string() }),
}, async (args) => {
  // Your logic here
  return { result: "Done!" };
});

The system handles everything else automatically:

Tool registration
Input validation
Connection management
Error handling
Logging

This means you can spend your time building tools that solve real problems instead of wrestling with MCP infrastructure.

Built for Teams, Not Just Demos

I built this because more and more clients are needing something production-ready:

Robust error handling that gives useful feedback
A system that scales from simple tools to complex integrations
Performance that holds up when multiple people are using it
An easy learning curve that anyone with Cursor can add to.

Deploy in Seconds, Not Days

Getting started is incredibly simple:

Hit the "Deploy on Railway" button for instant cloud setup
Or run pnpm install && pnpm dev locally for development

This instant deployment means you can start building tools right away without worrying about infrastructure.

What We're Building With This

My team is using this boilerplate to create tools that make our lives easier. Here are a few example ideas to get started:

Development tools that connect everything
- Pulling Jira ticket details into our AI workflows
- Creating branches with our naming conventions
- Sending Slack updates about our progress
Design and documentation flows
- Accessing Figma designs directly in our editor (CSS etc)
- Pulling in documentation as needed
- Updating docs as we work
Team communication made simpler
- Sending messages to teammates via Slack / Email without context switching
- Asking questions to specific teams for the bot to pick up later
- Creating automations for routine communication
- Storing new patterns and documents in Confluence.
Custom workflows for our specific needs
- Building tools that are unique to our company
- Connecting to our internal systems securely
- Making them available across all our AI interfaces

Getting Started Is Incredibly Easy

One of my main goals was to make this as accessible as possible:

Set up your server:
- One click with Railway
- Or run locally with a single command
Connect your AI tools:
- Point your AI at the /sse endpoint
- Start using your custom tools immediately
Create what you need:
- Add tools in src/modules/tools/
- Create prompts in src/modules/prompts/
- Everything gets registered automatically

Why Build Your Own?

There are great hosted solutions like Smithery.ai out there. They're excellent services! But there are real advantages to running your own MCP server:

Complete control - You decide what runs and what data is accessed
Custom integration - Connect directly to your internal systems
Learning opportunity - Understand how MCPs work from the inside
Independence - No reliance on third-party services

Join the MCP Ecosystem

I built this boilerplate because I believe in the potential of MCP to transform how we work with AI. By making it easier to create and deploy MCP servers, I hope more teams will build useful tools and share them with the community.

The current AI landscape is exciting but fragmented. MCPs provide a standardized way for models to interact with the world, and I want to help accelerate that adoption.

Check out the repo at github.com/dougwithseismic/withseismic-mcp and start building. I'd love to see what you create with it.

Doug Silkstone is the founder of WithSeismic. Find more at WithSeismic.com.