Europe 2026
Bending a Public MCP Server Without Breaking It — Nimrod Hauser, Baz
Overview
This talk explores strategies for effectively integrating and managing third-party tools within AI agent workflows, specifically focusing on Multi-Call Protocol (MCP) servers. It addresses the common challenges of generic tool descriptions, unexpected agent behavior, and potential performance degradation or security risks when using off-the-shelf tools. The presentation outlines a framework of five best practices to tailor these tools for specific use cases, ensuring agents perform as intended and applications remain stable.
Who should watch
- AI Engineers working with agentic systems.
- Product Managers evaluating agent capabilities.
- Builders integrating third-party tools into AI applications.
- Developers encountering unpredictable agent behavior with external tools.
- Anyone seeking to improve the reliability and performance of AI agents.
Key takeaways
- Third-party tools, while powerful, often have generic descriptions that can lead to agent misinterpretation and suboptimal performance.
- Curating tools by excluding unnecessary ones can reduce the agent's cognitive load and simplify decision-making.
- Wrapping tools with custom, detailed descriptions tailored to the specific use case significantly enhances agent understanding and behavior.
- Implementing deterministic guardrails is crucial for sensitive operations, preventing agents from performing unintended or risky actions.
- Composing new, specialized tools from existing ones allows for more nuanced control and better alignment with specific workflow needs.
- Treating certain tools as deterministic functions and executing them outside the agentic flow can be more reliable for complex or sensitive initial setup steps like authentication.
- The effectiveness of agentic tools relies on a continuous process of experimentation and adaptation to mold them to the specific application's requirements.
Notable quotes
*Third-party tools have their challenges. First and foremost, they might cause our agents to behave unexpectedly.*
*Sometimes, there are aspects of your tasks that are just too sensitive to leave at the hands of the agents.*
*At the end of the day, whatever gets our application to work as we want it, that's what we need to use.*
Unofficial community note. Prefer the recording for nuance.