AI & Machine Learning3 min read654 words

What Is Model Context Protocol (MCP)? A Practical Guide

MCP is an open standard that lets an AI model call your tools and read your data through one interface instead of a bespoke integration per vendor. Here is what it is, what it is not, and when it earns its place.

JL

Jishu Labs

Before MCP, wiring a model to your systems meant writing the same integration repeatedly — once for each model vendor, in each vendor's function-calling dialect. Swap the model and the work was largely thrown away. Model Context Protocol replaces that with one interface both sides agree on.

What MCP actually is

MCP is an open protocol that standardises how an application supplies context and capabilities to a language model. A **server** exposes tools, resources and prompts; a **client** — the model's host application — discovers and calls them. The model does not talk to your database. It asks the host to call a tool, and the host decides whether to allow it.

The useful mental model is a driver interface. USB did not make devices work; it made one plug work for many devices. MCP does not make your model smarter; it makes one integration work across hosts.

  • Tools — functions the model can invoke, with typed inputs and outputs.
  • Resources — data the host can read and place into context, addressed by URI.
  • Prompts — reusable templates a user can invoke deliberately.

Why it stopped being optional

MCP was donated to the Linux Foundation's Agentic AI Foundation in December 2025, with AWS, Google, Microsoft, Cloudflare and Bloomberg among the founding members. That matters more than any adoption figure: competing vendors agreeing on one protocol is what turns a good idea into an assumption.

The usage followed. Reported SDK downloads went from roughly 100,000 at launch to tens of millions per month, and the July 2026 specification revision moved the protocol toward a stateless architecture, which removes the session-affinity problem that made horizontal scaling awkward.

What a server looks like

typescript
// A minimal MCP server exposing one tool.
server.tool(
  'get_order_status',
  { orderId: z.string() },
  async ({ orderId }) => {
    const order = await db.orders.findUnique({ where: { id: orderId } });
    if (!order) return { content: [{ type: 'text', text: 'No such order.' }] };
    return { content: [{ type: 'text', text: `Status: ${order.status}` }] };
  }
);

Note what the tool does not do: it does not decide whether the caller is allowed to see that order. Authorisation belongs in your service, not in the tool description. A model can be argued into calling anything.

When MCP is the wrong choice

  • One model, one integration, no plans to change. A direct function-calling implementation is less indirection.
  • The action is not safe to expose. Anything irreversible — payments, deletions, outbound email — needs a human confirmation step regardless of protocol.
  • Latency is the product. Every hop costs. A tight retrieval loop may be better served by putting the data in context directly.

Treat every tool as an untrusted caller

MCP standardises *how* a model reaches your systems, not *whether it should*. Tool descriptions are model-visible text and can be manipulated by anything that reaches the context window. Enforce authorisation, rate limits and audit logging in the service behind the tool — never in the tool's description.

Where to start

Pick one read-only tool that answers a question your team asks often — order status, deployment state, on-call owner. Ship it behind existing authorisation. Read-only first means the worst failure is a wrong answer rather than a wrong write, which is the right way to learn where your context boundaries actually are.

Frequently Asked Questions

Is MCP tied to one AI vendor?

No. It was created at Anthropic and donated to the Linux Foundation's Agentic AI Foundation in December 2025, with AWS, Google, Microsoft, Cloudflare and Bloomberg as members. Multiple hosts implement it.

Does MCP replace RAG?

No. RAG is about retrieving relevant text; MCP is about calling tools and reading resources through a standard interface. An MCP server can expose a retrieval tool, but the two solve different problems.

Do I need MCP to build an agent?

No. You can implement function calling directly against one model's API. MCP earns its keep when you have several tools, several hosts, or expect to change models.

References

  1. Model Context Protocol — 2026-07-28 SpecificationModel Context Protocol
  2. The 2026 MCP RoadmapModel Context Protocol
  3. 2026: The Year for Enterprise-Ready MCP AdoptionCData
JL

About Jishu Labs

Jishu Labs is a software development company founded in 2016. We build custom software, AI/ML systems, and full-stack web and mobile applications for clients, and we make eight AI tools for software teams.

Related Articles

AI & Machine Learning3 min read

Small Language Models vs Frontier Models: A Cost Framework

Serving a 7B model is roughly 10-30x cheaper than a frontier model for tasks where accuracy is equivalent. The engineering question is which tasks those are, and how to find out without guessing.

Jishu Labs

July 28, 2026

AI & Machine Learning3 min read

What Is an AI Memory Layer?

Chat history is not memory. A memory layer is durable, retrievable state about decisions, preferences and facts that survives past the context window. What belongs in one, and what should stay in a log.

Jishu Labs

July 27, 2026

AI & Machine Learning3 min read

RAG in 2026: When You Still Need It, When You Don't

Long context windows and better tool use took work away from retrieval-augmented generation. RAG did not become obsolete — its job got narrower. A decision framework for when to retrieve, when to load, and when to call a tool.

Jishu Labs

July 23, 2026

Ready to Build Your Next Project?

Let's discuss how our expert team can help bring your vision to life.

AI Tools,
Built
End-to-End

Ready to Get Started?

Get consistent results. Collaborate in real-time.
Build Intelligent Apps. Work with Jishu Labs.

SCHEDULE MY CALL