docs/building-with-ai.mdx
We provide multiple tools to help AI coding assistants write correct Trigger.dev code. Use one or all of them for the best developer experience.
<Steps> <Step title="Install the MCP Server"> Give your AI assistant direct access to Trigger.dev tools — search docs, trigger tasks, deploy projects, and monitor runs. Works with Claude Code, Cursor, Windsurf, VS Code (Copilot), and Zed.npx trigger.dev@latest install-mcp
Learn more → </Step>
<Step title="Install Skills"> Portable instruction sets that teach any AI coding assistant Trigger.dev best practices: writing tasks, realtime frontends, and `chat.agent` AI agents. They install with the CLI into Claude Code, Cursor, VS Code (Copilot), and AGENTS-compatible tools such as Codex via `.agents/skills/`, and draw their API guidance from a version-pinned reference shipped in `@trigger.dev/sdk`.npx trigger.dev@latest skills
Learn more → </Step>
</Steps>Skills and the MCP server do different jobs and work best together. Here's how they compare:
| Skills | MCP Server | |
|---|---|---|
| What it does | Drops skill files into your project that teach Trigger.dev patterns | Runs a live server your AI connects to |
| Installs to | .claude/skills/, .cursor/skills/, .github/skills/, .agents/skills/ | mcp.json, ~/.claude.json, etc. |
| Updates | Re-run npx trigger.dev@latest skills, or auto-prompted on trigger dev | Always latest (uses @latest) |
| Best for | Teaching patterns and best practices | Live project interaction (deploy, trigger, monitor) |
| Works offline | Yes | No (calls Trigger.dev API) |
Our recommendation: Install both. Skills teach your AI how to write Trigger.dev code; the MCP Server lets it do things in your project.
If you prefer a lightweight/passive approach, paste the snippet below into a context file at the root of your project. Different AI tools read different files:
| File | Read by |
|---|---|
CLAUDE.md | Claude Code |
AGENTS.md | OpenAI Codex, Jules, OpenCode |
.cursor/rules/*.md | Cursor |
.github/copilot-instructions.md | GitHub Copilot |
CONVENTIONS.md | Windsurf, Cline, and others |
Create the file that matches your AI tool (or multiple files if your team uses different tools) and paste the snippet below. This gives the AI essential Trigger.dev context without installing anything.
<Accordion title="Copy the snippet"># Trigger.dev rules
## Imports
Always import from `@trigger.dev/sdk` — never from `@trigger.dev/sdk/v3` or use the deprecated `client.defineJob` pattern.
## Task pattern
Every task must be exported. Use `task()` from `@trigger.dev/sdk`:
```ts
import { task } from "@trigger.dev/sdk";
export const myTask = task({
id: "my-task",
retry: {
maxAttempts: 3,
factor: 1.8,
minTimeoutInMs: 500,
maxTimeoutInMs: 30_000,
},
run: async (payload: { url: string }) => {
// No timeouts — runs can take as long as needed
return { success: true };
},
});
```
## Triggering tasks
From your backend (Next.js route, Express handler, etc.):
```ts
import type { myTask } from "./trigger/my-task";
import { tasks } from "@trigger.dev/sdk";
// Fire and forget
const handle = await tasks.trigger<typeof myTask>("my-task", { url: "https://example.com" });
// Batch trigger (up to 1,000 items)
const batchHandle = await tasks.batchTrigger<typeof myTask>("my-task", [
{ payload: { url: "https://example.com/1" } },
{ payload: { url: "https://example.com/2" } },
]);
```
### From inside other tasks
```ts
export const parentTask = task({
id: "parent-task",
run: async (payload) => {
// Fire and forget
await childTask.trigger({ data: "value" });
// Wait for result — returns a Result object, NOT the output directly
const result = await childTask.triggerAndWait({ data: "value" });
if (result.ok) {
console.log(result.output); // The actual return value
} else {
console.error(result.error);
}
// Or use .unwrap() to get output directly (throws on failure)
const output = await childTask.triggerAndWait({ data: "value" }).unwrap();
},
});
```
> Never wrap `triggerAndWait` or `batchTriggerAndWait` in `Promise.all` — this is not supported.
## Error handling
```ts
import { task, retry, AbortTaskRunError } from "@trigger.dev/sdk";
export const resilientTask = task({
id: "resilient-task",
retry: { maxAttempts: 5 },
run: async (payload) => {
// Permanent error — skip retrying
if (!payload.isValid) {
throw new AbortTaskRunError("Invalid payload, will not retry");
}
// Retry a specific block (not the whole task)
const data = await retry.onThrow(
async () => await fetchExternalApi(payload),
{ maxAttempts: 3 }
);
return data;
},
});
```
## Schema validation
Use `schemaTask` with Zod for payload validation:
```ts
import { schemaTask } from "@trigger.dev/sdk";
import { z } from "zod";
export const processVideo = schemaTask({
id: "process-video",
schema: z.object({ videoUrl: z.string().url() }),
run: async (payload) => {
// payload is typed and validated
},
});
```
## Waits
Use `wait.for` for delays, `wait.until` for dates, and `wait.forToken` for external callbacks:
```ts
import { wait } from "@trigger.dev/sdk";
await wait.for({ seconds: 30 });
await wait.until({ date: new Date("2025-01-01") });
```
## Configuration
`trigger.config.ts` lives at the project root:
```ts
import { defineConfig } from "@trigger.dev/sdk/build";
export default defineConfig({
project: "<your-project-ref>",
dirs: ["./trigger"],
});
```
## Common mistakes
1. **Forgetting to export tasks** — every task must be a named export
2. **Importing from `@trigger.dev/sdk/v3`** — this is the old v3 path; always use `@trigger.dev/sdk`
3. **Using `client.defineJob()`** — this is the deprecated v2 API
4. **Calling `task.trigger()` directly** — use `tasks.trigger<typeof myTask>("task-id", payload)` from your backend
5. **Using `triggerAndWait` result as output** — it returns a `Result` object; check `result.ok` then access `result.output`, or use `.unwrap()`
6. **Wrapping waits/triggerAndWait in `Promise.all`** — not supported in Trigger.dev tasks
7. **Adding timeouts to tasks** — tasks have no built-in timeout; use `maxDuration` in config if needed
We also publish machine-readable documentation for LLM consumption:
These follow the llms.txt standard and can be fed directly into any LLM context window.