Author an app

Write an app in plain TypeScript: a provider, some queries and mutations, the approvals that guard them, stored data, and the call that deploys it.

An app is TypeScript source with a default defineApp export from the apps package. The host supplies that package. One index.ts is enough; you only need a package.json if you add a third-party dependency.

Write ordinary async TypeScript. You do not import Effect and you do not import the Executor SDK. The framework validates input and supplies the selected accounts.

A first app

import { query, defineApp, object, string } from "apps";

const Greet = object({ name: string().default("world") });

export default defineApp(
  { accounts: {} },
  {
    name: "Hello",
    queries: {
      greet: query(
        { description: "Greet someone by name", input: Greet },
        async (_ctx, { name }) => ({ message: `Hello, ${name}!` }),
      ),
    },
  },
);

defineApp(requirements, definition) takes what the app needs and what it offers. query(options, handler) and mutation(options, handler) declare the operations. Options are description, input, and optionally output and approval.

The schema helpers are object, string, number, boolean, array, record, json and literal. Use .optional() for a field that may be absent and .default(value) for a default. object strips properties you did not declare. Infer<typeof Input> gives you the parsed input type.

Using an account

Declare a provider, require it, and read it off the context.

import {
  query,
  type QueryContext,
  array,
  decodeJson,
  defineApp,
  defineProvider,
  object,
  secrets,
  string,
} from "apps";

const vercel = defineProvider({
  name: "Vercel",
  auth: {
    apiKey: secrets({
      label: "API token",
      fields: object({ token: string({ minLength: 1 }) }),
    }),
  },
});

const requirements = { accounts: { vercel } };
type Context = QueryContext<typeof requirements>;

const Projects = object({ projects: array(object({ id: string(), name: string() })) });

const listProjects = query(
  { description: "List projects for the selected Vercel account", input: object({}) },
  async ({ accounts, fetch }: Context) => {
    const response = await fetch("https://api.vercel.com/v9/projects?limit=10", {
      headers: { Authorization: `Bearer ${accounts.vercel.fields.token}` },
    });
    return decodeJson(response, Projects);
  },
);

export default defineApp(requirements, { name: "Vercel", queries: { listProjects } });

accounts.vercel.fields holds exactly the fields the method declared. The token arrives per call, from whichever account the app has selected. Nothing in the source names an account, so the same code serves a work app and a personal one.

decodeJson(response, schema) checks the HTTP status and parses the body. Use oauth2({ discover: "..." }) instead of secrets for a browser sign-in; the access token then arrives as accounts.<slot>.fields.access_token.

For a slot that takes several accounts, use provider.many(). The context gives you a list.

Queries and mutations

Both become tools. The difference is what they may do to the app’s own stored data.

A query gets a read-only view.
A mutation writes in a transaction that commits only after its output validates.

External calls are allowed from both, and external effects are never rolled back.

Approvals

Approval is declared on the operation.

import { always, never } from "apps/operations/approval";

// in the operation options:
approval: always(),   // ask a person first
approval: never(),    // run without asking

A function gets the real decision. It receives the tool name, the decoded input and an abort signal, and returns approved, denied or user-approval. Annotate a shared one with Approval<Input> to reuse it. For an imported operation, wrap it: withApproval(operation, policy).

There is no app-level approval setting, and nothing is inferred from an operation looking read-only. See Tools and approvals.

Asking for input mid-tool

await ctx.elicit({
  mode: "form",
  message: "Name this result",
  requestedSchema: {
    type: "object",
    properties: { name: { type: "string" } },
    required: ["name"],
  },
});

The question reaches the person the same way an approval does.

Importing an existing service

You do not have to write handlers to wrap an API.

import { defineApp } from "apps";
import { mcpOperations } from "apps/mcp";

export default defineApp({ accounts: {} }, async ({ signal }) => ({
  name: "DeepWiki",
  ...(await mcpOperations({
    url: "https://mcp.deepwiki.com/mcp",
    ...(signal === undefined ? {} : { signal }),
  })),
}));

The helpers are mcpOperations from apps/mcp, stdioOperations from apps/mcp/stdio, graphqlOperations from apps/graphql and openapiOperations from apps/openapi. Each returns { queries, mutations }. An operation that cannot be classified becomes a mutation.

The dashboard’s Custom app form does the same thing from a URL, for MCP, GraphQL and OpenAPI, without writing any source.

Storing data

Declare a database and the app gets tables.

import { defineDatabase, object, string, table } from "apps";

const database = defineDatabase({
  messages: table({ mailbox: string(), subject: string() }).index("by_mailbox", ["mailbox"]),
});

const requirements = { accounts: {}, database };

Query it through db on the context, with withIndex, order, and a terminal: first, take, collect, count or paginate({ cursor, numItems }). A by_creation index exists without being declared.

Reads are bounded: 5,000 rows scanned, 1,000 returned, 4 MiB per invocation. collect and count fail rather than truncate, so you notice. A mutation allows 1,000 writes.

Shipping instructions

An app can carry its own guidance for agents. Put it beside the source:

index.ts
skills/
- triage/
  - SKILL.md
  - references/
    - examples.md

---
name: triage
description: Search cached messages before fetching more history.
---

Read [examples](references/examples.md), then discover this app's queries.

An agent reads it with the skills tool. Skill text never grants a permission; the approval still decides.

Deploying

Deployment is an operation, not a command. Have your agent call it through the MCP endpoint.

return await tools.executor.mutations.apps_deploy({
  path: { organization: "<approved-organization-id>" },
  body: {
    name: "Hello",
    files: [{ path: "index.ts", content: "<contents of index.ts>" }],
  },
});

return await tools.executor.mutations.apps_deploy({
  body: {
    owner: "my-project",
    name: "Hello",
    files: [{ path: "index.ts", content: "<contents of index.ts>" }],
  },
});

The app activates only after the build succeeds, and earlier deployments are retained; see Apps and deployments. Then connect an account for each requirement and call it in a new execute:

return await tools["<app-slug>"].queries.greet({ name: "Ada" });

Discovery is not cached across a change, so run tools.search again after you deploy.

There is no executor deploy command. The Executor CLI starts the local server; it does not deploy app source.

Update a hosted app

Read the current source with apps_source. Submit the complete edited file set through apps_update, with expectedDeployment set to the source response’s id. The app keeps its ID and stored data. If another deployment won the race, read the source again before retrying. Use apps_activate to select a retained deployment; activation does not roll back stored data.

Open an app UI

Include ui/index.html and its browser scripts and styles in the deployment. The host builds and activates them with the server code. No separate publish step is needed.

For hosted apps, discover appUi_location through tools.search and call it:

return await tools.executor.queries.appUi_location({
  path: { organization: "<approved-organization-id>", app: "<app-id>" },
});

Open the returned url in a browser. Executor Cloud uses https://<app-slug>--<org-slug>.executor.website; self-host uses its configured app domain. Use the returned URL instead of guessing it. A null URL means the app has no UI or the host has no app domain configured.

App pages are private. Opening the link starts browser sign-in through Executor. A successful tool call or a 403 from a guessed address does not verify that the UI renders. Check the actual page before reporting it as working.

What is coming later

Scheduled and background work.
Calling one app from another.
Authoring general HTTP endpoints.
Stored-data schema migration between deployments.
A local executor dev loop.