One Flexible Tool Beats a Hundred Dedicated Ones

2026-05-18 · Source: Towards Data Science · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Advanced, medium

Summary

At the start of 2026, the standard approach for LLM agents interacting with systems like GitHub, Jira, or Postgres involved installing a Multi-Component Protocol (MCP) server, which exposes a menu of dedicated tools. However, this article argues that while MCP offers a good onboarding experience, it is often the wrong shape for real workloads. The core thesis is that a Command Line Interface (CLI) approach, which provides agents with one flexible tool, is superior with today's advanced models. The CLI allows agents to compose operations, manage multiple environments efficiently, and chain queries without bloating the context window, unlike MCP's dedicated tools that increase context cost with features and environments. The article demonstrates this through examples comparing Neo4j MCP server with `neo4j.sh` CLI for tasks like querying across environments, chaining queries, and piping data between different CLIs.

Key takeaway

For AI Engineers designing LLM agent interactions with external systems, consider adopting a CLI-centric approach over traditional MCP servers. Your agents will benefit from reduced token costs and enhanced operational flexibility, especially when managing multiple environments or chaining complex data flows. Implement robust sandboxing and allowlisting to mitigate the increased blast radius associated with terminal access, ensuring secure and efficient agent deployments.

Key insights

Flexible CLI tools outperform dedicated MCP tools for LLM agents due to improved compositionality and reduced context bloat.

Principles

• Fewer, more expressive tools are better for agents.
• Agents excel at composing operations with flexible tools.
• Context window bloat is a critical performance factor.

Method

Transition from dedicated MCP tools to flexible CLI tools for LLM agent interactions. This involves providing agents with a terminal-like interface that supports composition via pipes, variables, and loops, allowing data to flow outside the agent's context.

In practice

• Use `for` loops for multi-environment queries.
• Pipe CLI outputs (`|`) to chain operations.
• Employ `jq` for data reshaping between CLIs.

Topics

• LLM Agents
• Multi-Component Protocol
• Command-Line Interface
• Context Window Optimization
• Data Pipelining

Code references

• neo4j/mcp

Best for: AI Engineer, Machine Learning Engineer, AI Architect

Related on AIssential

• Whitepaper Companion Podcast - Agent Tools & Interoperability with MCP — MCP standardizes LLM tool integration, enabling real-world actions and fostering a modular, scalable AI agent ecosystem.
• Every Software Just Became Agent-Native — AI agents are gaining universal software control and self-improvement capabilities through CLI harnesses and skill optimization.
• The Infrastructure Behind Making Local LLM Agents Actually Useful — Local LLM agents require deliberate infrastructure for performance and long-session stability beyond basic model serving.
• Why Google Workspace CLI is a Big Deal — CLIs are becoming the preferred interface for AI agents due to their low friction and deterministic output.
• The Sequence Opinion #848: The Agent’s Hands: CLI or MCP? — Agentic system efficacy hinges on the interface connecting the language model to external tools, not solely on the model itself.
• Accessing watsonx Orchestrate from Bob via CLI — Using agentic skills with CLIs offers advantages over MCP tools for AI development agents like IBM Bob.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards Data Science.