Skip to main content
bash TV

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

AI Engineer

1.2K views9 Oct 2026

YouTube

Airbyte built both an MCP server and a CLI so agents can use its data. Here's what each one is actually good for. Pedro Lopez, software engineer at Airbyte, shares lessons from building two ways for AI agents to reach company data through Airbyte's Context Store, a search-optimized index of tools like Zendesk, Stripe and HubSpot. Both are thin interfaces over one platform, built from OpenAPI specs. For the MCP server, he explains why to keep the tool count small and use progressive discovery, why OAuth is the only realistic way to sign in, why users expect long-lived sessions, and why URL elicitation isn't ready across clients yet. For the CLI, he explains designing for agents instead of humans: JSON in and JSON out, consistent noun-verb commands, skills that ship with the tool, and no interactive prompts. He closes with when to choose MCP and when to choose a CLI. In this talk: • Keeping MCP tool counts small with progressive discovery • OAuth, long sessions, and elicitation in real MCP clients • Designing a CLI for agents: JSON in, JSON out, consistency, skills • When to choose MCP, and when to choose a CLI SPEAKER Pedro Lopez, Software Engineer, Airbyte LinkedIn: https://www.linkedin.com/in/pedroslopez/ GitHub: https://github.com/pedroslopez Website: https://pedroslopez.me/ LINKS Airbyte Context Store: https://airbyte.com/agentic-data/context-store Airbyte agent MCP docs: https://docs.airbyte.com/ai-agents/interfaces/mcp/ Airbyte Agent CLI: https://airbyte.com/agent-cli Airbyte on X: https://x.com/AirbyteHQ CHAPTERS 0:00 Intro 0:32 From data movement to AI agents 1:02 The Context Store 2:12 What the interfaces expose 3:07 Thin interfaces on one platform 3:57 The MCP server 4:52 Demo: MCP in ChatGPT 5:52 Keep the tool count small 6:52 OAuth is the way 7:56 Users expect long sessions 8:31 Elicitation isn't ready yet 10:11 The CLI 10:26 Demo: the CLI 11:11 Designing CLIs for agents 11:31 JSON in, JSON out 12:36 Consistency 13:51 Ship skills with the CLI 15:06 No prompts: auth for agents 16:01 MCP vs CLI Recorded at the AI Engineer World's Fair 2026 in San Francisco. Subscribe for more talks from the engineers building with AI. AI Engineer: https://ai.engineer YouTube: https://www.youtube.com/@aiDotEngineer X: https://x.com/aiDotEngineer LinkedIn: https://www.linkedin.com/company/aidotengineer/ #MCP #AIAgents #AIEngineer

Join the discussion

Sign in to join the discussion

Sign in