Kody

Run one ephemeral ESM module. Discover the capability with search and adapt that entity's execute snippet. Prefer a package over rewriting helpers. Project large results before returning (e.g. { id, subject, snippet }). Same user and `code` graph reuse one isolate for the UTC day — vary args via `params`. import { kody } from 'kody:runtime' export default async function main(params) { return await kody.capability_id(params) }

execute

Find built-in capabilities, official guides, saved packages, integrations, connected MCP servers, and secret references (metadata only) before `execute`. **query** — compact ranked markdown + structured matches. Empty or broad queries return a domain index; search again with a more specific query. Domain ids appear on capability hits. Connected MCP servers appear as server hits (name, instructions); do not expect every remote tool in unscoped results. **entity: "{id}:{type}"** — detail for one hit (`capability` | `guide` | `integration` | `mcp-server` | `package` | `secret`), or 1–10 refs. Guide detail is the full markdown when it fits the response budget; oversized guides return a table of contents. Open a heading with `{id}:guide#{slug}` (for example `package_subscriptions:guide#repo.pushed`). Capability detail includes an execute snippet. MCP server detail lists discovered tools; after that, call `kody.mcp["name"].tool_name(args)`. Example arguments: - `{ "query": "send a message" }` - `{}` - `{ "query": "send a message", "domain": "email" }` - `{ "domain": "jobs" }` - `{ "entity": "package_authoring:guide" }` https://github.com/kentcdodds/kody/blob/main/docs/use/search.md

search