---
title: "Native Automations with SDK Primitives"
description: "Run app-owned schedulers while using integrate-sdk for OAuth, tool execution, and step results"
---
Production apps like [Individu](https://github.com/integratedotdev/individu) often outgrow SDK-managed trigger CRUD but still rely on integrate-sdk for the hard parts: OAuth tokens, MCP tool execution, and multi-step callbacks.

## When to use what

| Pattern | Use when |
|---------|----------|
| **SDK triggers** (`triggers.onComplete`, `serverClient.trigger.*`) | You want scheduler registration + CRUD handled by the SDK |
| **Native orchestration** | You own scheduling (cron, queues, Vercel cron) but call MCP tools server-side |
| **Hybrid** | Legacy SDK triggers + new native automations; share `onComplete` for multi-step flows |

## Server-side tool execution

Use the public `callTool` API (not `_callToolByName`):

```typescript
import { serverClient } from "@/lib/integrate";
import { parseMCPToolResult } from "integrate-sdk/server";

const raw = await serverClient.callTool("github_list_repos", { per_page: 10 }, {
  context: { userId },
});

const result = parseMCPToolResult(raw);
if (!result.ok) {
  throw new Error(result.error);
}
```

`parseMCPToolResult` normalizes MCP `isError`, `success: false`, `structuredContent`, and text `content` blocks.

## Multi-step orchestration (`onComplete`)

Wire SDK trigger callbacks when steps should chain:

```typescript
createMCPServer({
  // ...
  triggers: {
    onComplete: async (ctx) => {
      // ctx.request.steps: normalize with parseMCPToolResult per step
      // return { hasMore, nextStep } for chained execution
      return { hasMore: false };
    },
    getCallbackUrl: () => process.env.APP_URL!,
  },
});
```

Types: `StepResult`, `CompleteCallbackContext`, `CompleteResponse` from `integrate-sdk`.

## Template variables between steps

Resolve outputs from prior steps in your orchestrator (see Individu's `lib/automations/template.ts` pattern). Keep templates in your app; the SDK provides execution + result parsing only.

## Performance for automation agents

Reuse cached AI tools across chat and automation paths:

```typescript
import { getVercelAITools } from "integrate-sdk/server";
import { createMemoryToolDiscoveryCache } from "integrate-sdk/server";

const cache = createMemoryToolDiscoveryCache();

const tools = await getVercelAITools(serverClient, {
  context: { userId },
  connectedOnly: true,
  mode: "code",
  cache: { cache, ttlMs: 5 * 60 * 1000 },
});
```

Invalidate on OAuth connect/disconnect via `createToolDiscoveryCacheInvalidator(cache)` in your database adapter `onTokenChange` hook.