Inngest Plugin for Claude Code

The official Inngest plugin for Claude Code. One install, and Claude Code knows how to build reliable durable functions, design event-driven workflows, configure flow control, and stream realtime updates with Inngest.

Beta: v0.1.0 is the first public beta. We’d love your feedback — open an issue, drop into our Discord, or ping @inngest on socials.

What’s included

• 7 skills covering setup, events, durable functions, steps, flow control, middleware, and realtime — Claude Code loads the right one automatically based on what you’re building.
• MCP server for the local Inngest dev server. Claude Code can inspect runs, events, and function state on your machine while you work.
• Eval harness so you can verify the skills are actually steering Claude Code on your codebase (and contribute new prompts back).
• More on the way — commands, agents, and migrators are tracked in ROADMAP.md.

Installation

Claude Code marketplace

/plugin marketplace add inngest/inngest-claude-code-plugin
/plugin install inngest@inngest-claude-code-plugin

Local clone

git clone https://github.com/inngest/inngest-claude-code-plugin.git
cd your-project/
claude --plugin-dir /path/to/inngest-claude-code-plugin

Quick start

1. Install the plugin (above).

2. Start your Inngest dev server in your project:

   npx inngest-cli@latest dev
   

3. Open Claude Code in your project and ask it something Inngest-shaped:

   "Add a durable function that sends a welcome email when a user signs up,
    retries on failure, and waits 24 hours before sending the second email."
   

   Claude Code picks up the relevant skill (inngest-durable-functions,
   inngest-steps, etc.), writes the function, registers it on your dev
   server’s /api/inngest route, and triggers a test event through the MCP
   server so you can watch it run.

Skills

Skill	What it covers	When it triggers
inngest-setup	SDK install, client config, serve endpoints (Next.js, Express, Hono, Fastify), connect-as-worker, dev server	Adding Inngest to a TypeScript project
inngest-durable-functions	Function config, triggers, step execution, idempotency, cancellation, error handling, retries, observability	Building functions that survive crashes, retry, run on schedule
inngest-steps	step.run, step.sleep, step.waitForEvent, step.invoke, step.ai, parallel + loops	Durable delays, human-in-the-loop, polling, memoization
inngest-events	Event schema, IDs for idempotency, fan-out patterns, system events	Designing event-driven workflows, decoupling services
inngest-flow-control	Concurrency, throttle, rate limit, debounce, priority, singleton, batching	Handling rate limits, deduping bursts, per-tenant fairness
inngest-middleware	Lifecycle, dependency injection, Sentry + encryption middleware, custom middleware	Cross-cutting concerns: logging, tracing, DI, encryption
inngest-realtime	v4 native realtime, channels, subscription tokens, useRealtime hook, SSE	Streaming workflow updates to a UI in real time

Dev server MCP

The plugin ships an .mcp.json that registers the local Inngest dev server’s MCP endpoint with Claude Code:

{
  "mcpServers": {
    "inngest-dev": {
      "type": "http",
      "url": "http://127.0.0.1:8288/mcp"
    }
  }
}

This lets the agent inspect runs, events, and function state on your local dev server while you’re working.

Port note: the URL is hardcoded to 8288, the dev server’s default. If 8288 is already in use, the dev server falls back to 8289+ — in that case, edit the url in .mcp.json to point at the active port. Run lsof -i :8288 (or check the dev server’s startup output) to find which port it bound to.

What you can do

Build a retry-safe webhook handler

"I'm getting Stripe webhooks. Wrap the handler in an Inngest function that
 retries on failure and dedupes on event ID."

→ Plugin generates the serve endpoint, the function with id set as the idempotency key, and the step.run blocks for each side effect.

Add concurrency limits to an OpenAI-heavy function

"This function calls OpenAI on every event. Add concurrency: max 5 in
 flight per user, throttle the org to 100/min."

→ Plugin writes the concurrency and throttle config keyed by user/org.

Stream agent tokens to a Next.js page in realtime

"Run this LLM agent as an Inngest function and stream its tokens to a
 React component."

→ Plugin sets up a typed channel, publishes from inside step.run, mints a subscription token via a server action, and writes the client component using the useRealtime hook.

Skills source of truth

The skills in this plugin are mirrored from inngest/inngest-skills — that’s where they’re authored and where skills.sh users install them. This plugin pulls them in via scripts/sync-skills.sh so the Claude Code experience stays in lockstep with the skills.sh experience.

Eval harness

Inside eval/ is a prompt catalog and runner you can use to verify the plugin is actually steering Claude Code on real Inngest-shaped tasks.

cd eval/runner
./run.sh

See eval/README.md for prompt format, judge config, and how to add your own.

Beta feedback

This is v0.1.0. We’re shipping early to learn from real usage:

• What’s missing: open a GitHub issue — even a one-liner helps.
• What’s broken: same place. Include the prompt and the skill that fired (or didn’t).
• What to build next: chime in on ROADMAP.md discussions.
• Real-time chat: Inngest Discord — there’s a #claude-code-plugin channel.

Roadmap

See ROADMAP.md for the v0.x → v1.0 plan: more skills, slash commands, agents, migrators between SDK majors, and the path to the official Anthropic plugin marketplace.

License

Apache 2.0. See LICENSE.