Sign in Subscribe

MCP only becomes powerful when you use the features that actually matter for agent performance

MCP only becomes powerful when you use the features that actually matter for agent performance
Photo by Luca Bravo / Unsplash

After working hands-on with dozens of AI agent architectures since 2022, it is clear to me that the Model Context Protocol (MCP) represents a major shift. MCP gives us a standard way to expose capabilities, tools, resources, sessions, and context directly to models. This is exactly what modern AI systems need to move beyond simple chat interfaces and into real operational workflows.

Here is the part most teams are missing.

If you use MCP as a simple API wrapper, you are not only misusing the protocol, you are also degrading the performance of your AI solution.

This is what context engineering teaches us.

The value of an MCP server is not in the function call itself. The value only appears when the output reaches the context window of the model and contributes to a better final answer.

Most MCP servers never get to that point. Instead they wrap basic business logic such as:

  • CRUD reads
  • static API calls
  • helpers that could live in a normal backend
  • prompt templates with no session state
  • thin glue around existing systems

When MCP is used this way, it becomes an expensive detour.
You increase latency. You increase complexity. You scatter context across places that do not help the model.
And the final LLM output does not improve.

For simple integrations, REST or GraphQL is usually the better choice. They are mature, well tooled, and designed for traditional application logic.

MCP only becomes powerful when you use the features that actually matter for agent performance:

  • bidirectional streaming
  • capability discovery the model can understand
  • long running sessions with state
  • structured resource access
  • incremental reads and deltas
  • model facing safety rules
  • multi agent workflows that share context in real time
💡
Note: Some of the features I describe (stateful sessions, streaming, multi-agent context sharing) are not enforced by the MCP specification, but they are enabled by MCP’s primitives, and are common in production-grade implementations. See source below.

When you use these features, the MCP server produces context the model can actually use. The LLM produces better answers. The agent chain becomes more reliable. The entire workflow becomes more predictable.

That is the real purpose of MCP.
Not more endpoints.
Not another RPC layer.
A way to deliver context to the model in a form that improves the final result.

The takeaway is simple.

If you only need CRUD or static logic, use REST or GraphQL.
If you need real agent capabilities and context aware behavior, use MCP.

APIs handle your data.
MCP handles your capabilities and your model context.

If you want to see how we are applying this for mid-market clients and beyond, reach out. I am happy to share what has worked and what has not.

Sources

Introducing the Model Context Protocol
The Model Context Protocol (MCP) is an open standard for connecting AI assistants to the systems where data lives, including content repositories, business tools, and development environments. Its aim is to help frontier models produce better, more relevant responses.
What is the Model Context Protocol (MCP)
The Model Context Protocol (MCP) was created to address a fundamental challenge in building agentic AI applications: connecting isolated large language models (LLMs) to the outside world. By default,

Elastic The Search AI Company
Specification - Model Context Protocol
Model Context Protocol
Why MCP Won
Learnings from Anthropic’s extraordinarily successful Launch and Workshop
Latent.Space
Security Risks of Agentic AI: A Model Context Protocol (MCP) Introduction
What is Model Context Protocol (MCP) and how could it be a target for someone trying to tamper with a system? This Agentic AI blog explores these questions.
Bitdefender SRLMartin Zugec
MCP is the new WWW - Scale Venture Partners
Over the past year, agentic AI has been loosely defined to mean anything from RAG to function calling and computer use. But what is an agent exactly? We now have an answer: agent = LLM + MCP. Primer For LLMs, MCP serves the same goal that web browsers accomplished for humans 30 years ago: it

Scale Venture PartnersJavier Redondo
ScaleMCP: Dynamic and Auto-Synchronizing Model Context Protocol Tools for LLM Agents
Recent advancements in Large Language Models (LLMs) and the introduction of the Model Context Protocol (MCP) have significantly expanded LLM agents’ capability to interact dynamically with external tools and APIs. However, existing tool selection frameworks do not integrate MCP servers, instead relying heavily on error-prone manual updates to monolithic local tool repositories, leading to duplication, inconsistencies, and inefficiencies. Additionally, current approaches abstract tool selection before the LLM agent is invoked, limiting its autonomy and hindering dynamic re-querying capabilities during multi-turn interactions. To address these issues, we introduce ScaleMCP, a novel tool selection approach that dynamically equips LLM agents with a MCP tool retriever, giving agents the autonomy to add tools into their memory, as well as an auto-synchronizing tool storage system pipeline through CRUD (create, read, update, delete) operations with MCP servers as the single source of truth. We also propose a novel embedding strategy, Tool Document Weighted Average (TDWA), designed to selectively emphasize critical components of tool documents (e.g. tool name or synthetic questions) during the embedding process. Comprehensive evaluations conducted on a created dataset of 5,000 financial metric MCP servers, across 10 LLM models, 5 embedding models, and 5 retriever types, demonstrate substantial improvements in tool retrieval and agent invocation performance, emphasizing ScaleMCP’s effectiveness in scalable, dynamic tool selection and invocation.
arXiv.orgElias Lumer