> Markdown version of [/videos/100202-mcp-doesn-t-suck-your-agent-does](https://www.wearedevelopers.com/videos/100202-mcp-doesn-t-suck-your-agent-does). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# MCP doesn’t suck — your agent does

The Model Context Protocol isn't broken, but your naive agent implementation is. Discover how combining MCP with traditional CLI workflows eliminates context bloat and secures your remote endpoints.

- **Speakers:** [Jan Curn](https://www.wearedevelopers.com/@jan-curn)
- **Event:** World Congress 2026 Europe
- **Published:** July 10, 2026
- **Duration:** 27:55
- **URL:** https://www.wearedevelopers.com/videos/100202-mcp-doesn-t-suck-your-agent-does

## Summary

The Model Context Protocol (MCP) promised to standardize how AI agents interact with external tools, but it quickly faced community backlash. Developers flagged severe context window bloat, wasted token spend, and "context rot." However, the core flaw isn't MCP itself, but naive agent implementations that blindly inject entire tool registries and raw execution results into prompt-time function calls. This frustration fueled a narrative that MCP is a flawed abstraction and traditional Command Line Interfaces (CLIs) are the superior alternative, largely because Large Language Models (LLMs) possess deep, natively trained proficiency in shell scripting.

Rather than treating the protocol and the terminal as mutually exclusive, combining them mitigates both context exhaustion and remote access vulnerabilities. While CLIs are excellent local interfaces—letting models pipe intermediate output between commands securely without polluting the context window—they fail as opaque, non-standardized remote endpoints. By leveraging efficient architectural patterns like progressive tool discovery and "code mode," agents can dynamically search for necessary tools at runtime and evaluate logic programmatically rather than relying on artificial tool-calling tokens.

To bridge this gap, Apify developed `mcpc`, an open-source universal CLI client that maps complex protocol operations to intuitive shell commands. With a simple `Bash()` call, AI agents can seamlessly self-discover tools, parse JSON, and handle newly standardized protocol features like asynchronous tasks and stateless sessions. This hybrid architecture offloads rigorous enterprise access controls and secure proxies to MCP's standardization, while enabling cutting-edge capabilities like XPRL2 agent payments. Ultimately, merging MCP's robust remote authentication with CLI's operational efficiency resolves the debate, yielding a context-saving, highly reliable execution environment for modern agents.

**Keywords:** model context protocol optimization, AI agent tool calling, progressive tool discovery, context window bloat, prompt-time function calls, code mode execution, bash scripting for LLMs, universal CLI client, asynchronous protocol tasks, context rot prevention, XPRL2 agent payments, secure remote access controls, stateless MCP sessions, LLM context efficiency, enterprise proxy configuration

## Chapters

1. **Overview of the Model Context Protocol and its rapid adoption** (00:03) — The fast-growing open-source protocol standardized AI agent interactions with remote and local tools.
1. **The backlash against MCP due to context window overload** (01:47) — Eagerly injecting tool definitions and operational outputs directly into the prompt causes significant token waste and context rot.
1. **Splitting context into sub-agents to manage token constraints** (04:53) — Routing discrete tasks into separate agent contexts reduces main prompt clutter but maintains overall cumulative token costs.
1. **Implementing progressive tool discovery for efficient prompt usage** (05:50) — Creating a single tool-search function limits context bloat by allowing intelligent agents to query capabilities dynamically.
1. **Using code mode to improve agent tool calling execution** (07:14) — Translating tool operations into code representations enables models to chain complex logic without polluting conversational token limits.
1. **Why command line interfaces excel as local agent tools** (09:40) — Natural proficiency with historical shell environments allows models to execute efficient multi-command pipelines without loading explicit tooling setups.
1. **The limitation of command line interfaces for enterprise control** (12:14) — Operating as unstandardized local black boxes prevents administrators from inspecting authentication mechanisms and remote server access appropriately.
1. **Combining standard protocols with CLI execution using the mcpc tool** (13:27) — Packaging sophisticated remote capabilities into an intuitive shell command enables native local interaction minus massive initial context payloads.
1. **Live demonstration of connecting mcpc to local file systems** (15:44) — Exploring local host configurations through native terminal output and strict JSON formats illustrates immediate technical integration methods.
1. **Handling remote server authentication and asynchronous workflow tasks** (18:03) — Initiating robust credential caching and tracking persistent background procedures highlights protocol optimizations for longer execution timelines.
1. **Evaluating connector performance and supporting advanced agent standards** (22:09) — Profiling token efficiency across different framework variations identifies cost savings while adopting modern enterprise payment primitives.
1. **Audience questions regarding massive tool filtering and enterprise permissions** (25:28) — Navigating enterprise environments with hundreds of remote actions requires structural gateways to enforce isolated sub-agent boundaries successfully.

## Related Moments

- [Core concepts and advantages of the Model Context Protocol](https://www.wearedevelopers.com/videos/100305-api-mcp-or-mcp-app-choosing-the-right-surface-for-ai-agents) (from "API, MCP or MCP App? Choosing the right surface for AI agents")
- [Standardizing agent interactions with the Web MCP proposal](https://www.wearedevelopers.com/videos/1805-wearedevelopers-live-modern-devops-for-iot-devices-and-more) (from "WeAreDevelopers LIVE - Modern DevOps for IoT Devices and More")
- [Origin and purpose of the Model Context Protocol](https://www.wearedevelopers.com/videos/100132-the-agent-interface-layer-protocols-tools-and-trust-boundaries) (from "The Agent Interface Layer: Protocols, Tools and Trust Boundaries")
- [Integrating Model Context Protocol for non-technical users](https://www.wearedevelopers.com/videos/100091-3-ways-to-rebuild-the-data-stack-for-agents) (from "3 Ways to Rebuild the Data Stack for Agents")
- [Evaluating MCP protocols versus CLI for enterprise agent architectures](https://www.wearedevelopers.com/videos/100092-headless-by-design-building-enterprise-systems-that-agents-can-actually-use) (from "Headless by Design: Building Enterprise Systems That Agents Can Actually Use")
- [Standardizing agent tool discovery with MCP](https://www.wearedevelopers.com/videos/100336-from-apis-to-mcp-enterprise-governance-registry-and-controls) (from "From APIs to MCP: Enterprise Governance, Registry, and Controls")

## Related Articles

- [Everything a Developer Needs to Know About MCP with Neo4j](https://www.wearedevelopers.com/magazine/604-everything-a-developer-needs-to-know-about-mcp-with-neo4j)
- [WebMCP: Empowering Agents as First-Class Citizens of the Web](https://www.wearedevelopers.com/magazine/696-webmcp-empowering-agents-as-first-class-citizens-of-the-web)
- [Dev Digest 210: AI Agents Are Go! Is MCP Dead? LLMs Crack Anonymity](https://www.wearedevelopers.com/magazine/709-dev-digest-210-ai-agents-are-go-is-mcp-dead-llms-crack-anonymity)
- [What is Agentic Programming and Why Should Developers Care?](https://www.wearedevelopers.com/magazine/625-what-is-agentic-programming-and-why-should-developers-care)

## Related Jobs

- [Senior Backend Developer — AI: MCP & Agent Engine](https://www.wearedevelopers.com/jobs/48297-senior-backend-developer-ai-mcp-agent-engine) at **basebox GmbH**
- [Principal Product Manager, Agent Platform](https://www.wearedevelopers.com/jobs/ext/277541-principal-product-manager-agent-platform) at **GitHub**
- [Senior Engineer, Infrastructure Platform](https://www.wearedevelopers.com/jobs/ext/328836-senior-engineer-infrastructure-platform) at **Intercom, Inc.**
- [Principal Field Architect - AI Agents](https://www.wearedevelopers.com/jobs/ext/1442858-principal-field-architect-ai-agents) at **Twilio**
- [Senior AI Agent Software Engineer (Go, Python) (m/f/x)](https://www.wearedevelopers.com/jobs/48277-senior-ai-agent-software-engineer-go-python-m-f-x) at **Dynatrace**
- [Engineer, Offensive Security Organization](https://www.wearedevelopers.com/jobs/ext/1992296-engineer-offensive-security-organization) at **Twilio**