Designing CLIs for Agents, Not Humans — Pedro Lopez, Airbyte

Gives concrete design rules for exposing a product to agents, including when a JSON-in/JSON-out CLI with shipped skills beats an MCP server, and the OAuth and session issues MCP clients still have.

• Airbyte's MCP server and CLI are both thin interfaces built from OpenAPI specs over one platform, its Context Store, a search-optimized index of tools like Zendesk, Stripe and HubSpot.
• For the MCP server, Lopez recommends keeping the tool count small with progressive discovery, calls OAuth the only realistic sign-in, notes users expect long-lived sessions, and says URL elicitation isn't ready across clients.
• For the agent-facing CLI: JSON in and JSON out, consistent noun-verb commands, skills that ship with the tool, and no interactive prompts, including for authentication.

• MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
• AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
• context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

• videoWhy Your Company Needs a Context Graph (and How to Build It) — Gil Feig, Merge
• videoSmall Models, Big Results: Training a Finance Agent for Under $500 — Charles Dickens, Snorkel AI
• videoFrom Context to Memory: Your Agents Need a Real Memory Layer — Anders Swanson, Oracle
• videoGeneration Is Cheap, Review Is Expensive: How to Stop Shipping AI Slop — Gabriel Martinez, G2i

• videoMCP Doesn't Suck. Your Agent Does. — Jan Čurn, ApifyAI Engineer
• videoRebuilding the web for agents — Liad Yosef, MCP AppsAI Engineer
• videoWhy We Deleted Our MCP Server and Rebuilt It — Abhi Arya, ReductoAI Engineer

Checking sign-in…

Loading comments…