Uploaded May 2026 | Updated September 2026, 16 hours ago
Docs became skills. API calls became MCP tools. But UI? Agents have been stuck rendering plain text while we spent years building components that explain what's actually happening in our products. Rhys Sullivan, founder of executor.sh, shows how to close that gap today — without waiting for the MCP apps spec to land.
0:00 Intro — the missing UI layer for agents
0:44 PostHog's in-product assistant demo
1:41 Calling render UI from a coding agent
2:47 Shipping your own React components to the agent
3:14 The four tools and the executor SDK
5:22 MCP apps in Claude Code
Rhys walks through a working pattern: convert your OpenAPI, GraphQL, or MCP server into a TypeScript SDK the agent can write against, expose your real React components (he drops in PostHog's time series chart), and give the agent a render UI tool that returns a deep link into your product. The result is the PostHog dashboard experience showing up inside a coding agent — interactive charts, tooltips, the works — without leaving the editor.
Along the way he points at Cloudflare's code mode write-up as the conceptual backbone (agents are good at writing code, so give them an environment to write it in), and shows a Gmail example where the agent filters response emails from OpenAI and Anthropic with a few lines of TypeScript. The close: when clients support MCP apps natively you embed directly; when they don't, a deep link is already enough to ship great experiences.
executor.sh: https://executor.sh
Docs became skills. API calls became MCP tools. But UI? Agents have been stuck rendering plain text while we spent years building components that explain what's actually happening in our products. Rhys Sullivan, founder of executor.sh, shows how to close that gap today — without waiting for the MCP apps spec to land.
0:00 Intro — the missing UI layer for agents
0:44 PostHog's in-product assistant demo
1:41 Calling render UI from a coding agent
2:47 Shipping your own React components to the agent
3:14 The four tools and the executor SDK
5:22 MCP apps in Claude Code
Rhys walks through a working pattern: convert your OpenAPI, GraphQL, or MCP server into a TypeScript SDK the agent can write against, expose your real React components (he drops in PostHog's time series chart), and give the agent a render UI tool that returns a deep link into your product. The result is the PostHog dashboard experience showing up inside a coding agent — interactive charts, tooltips, the works — without leaving the editor.
Along the way he points at Cloudflare's code mode write-up as the conceptual backbone (agents are good at writing code, so give them an environment to write it in), and shows a Gmail example where the agent filters response emails from OpenAI and Anthropic with a few lines of TypeScript. The close: when clients support MCP apps natively you embed directly; when they don't, a deep link is already enough to ship great experiences.
executor.sh: https://executor.sh
