Inngest MCP Server

The Inngest MCP Server connects your AI assistant to Inngest. Inngest is a workflow orchestration platform: you write functions made up of several steps that are triggered by events. Each execution of a function is called a run. Inngest keeps track of which steps have already finished and retries failed steps automatically. MCP, the Model Context Protocol, is an open standard that assistants such as Claude Code, Codex or Cursor use to call external tools. According to its own documentation, Inngest runs not one but two official MCP servers: a hosted one for Inngest Cloud and a local one built into the Inngest Dev Server. You can set up both side by side in the same client.

What is the Inngest MCP Server useful for?

Event-driven systems are hard to keep track of. One event triggers several functions, each function has several steps, and a bug often hides in the third step of a run that happened two hours ago. Normally you would click your way through the dashboard to find it. With the MCP server, the assistant queries the runs itself, reads the trace tree including the output of individual steps and connects what it finds to your code. Once a fix is in, it reruns the function, either from the start or from exactly the step that failed. Rerunning from a chosen step is what sets Inngest apart from many other job platforms. It saves time because expensive earlier steps, such as AI calls, do not have to run again.

Local server in the Dev Server

The Dev Server is Inngest's local development environment. You start it with npx inngest-cli@latest dev, and the MCP endpoint is then available at http://127.0.0.1:8288/mcp. No sign-in is required. The assistant sees your local apps, functions, events and runs, and it can send events, invoke functions and cancel or rerun runs. On top of that, the Dev Server ships with the Inngest documentation built in. Using the grep_docs, list_docs and read_doc tools, the assistant looks things up there instead of guessing from outdated training data. That is particularly handy when you write new functions.

Hosted server for Inngest Cloud

The cloud server at https://api.inngest.com/mcp works with your deployed apps. According to the documentation, it covers apps, functions, events, runs and traces, as well as environments, webhooks, sessions, experiments and Insights. Insights is Inngest's query interface. The assistant can run SQL queries there or have a plain-language question turned into SQL. It can also sync apps, create and archive environments, set up inbound webhooks and attach scores to runs, for example to evaluate AI experiments. The recommended sign-in is OAuth in the browser, where you pick the account, environment, permissions and session length. An API key passed as a bearer token works too; signing keys are not accepted.

Take care with tools that write

Many tools change data: sending events, invoking functions, cancelling or rerunning runs, syncing apps. The documentation explicitly advises caution, especially in production. It does not mention a dedicated read-only mode like some other servers offer. Your main safeguard is the OAuth session, where you grant only the permissions and environment you actually need. Tools that return event keys or signing keys deliberately do not work over OAuth at all.

Cost and licence

The MCP servers themselves cost nothing extra. The Dev Server is free to use. For the cloud server you need an Inngest account. According to the pricing page, there is a free plan with no credit card and a fixed monthly allowance of executions, plus paid Pro, Business and Enterprise plans with usage-based billing. Inngest's source code lives in the inngest/inngest repository under the Server Side Public License (SSPL), with a committed switch to Apache 2.0 after three years.

Who is the Inngest MCP Server for?

It pays off for teams that use Inngest for background jobs, multi-step workflows or AI agents and want to narrow down failures faster. Newcomers get the most out of the local server with its built-in documentation. For Claude Code there is also an official plugin that bundles the cloud server with matching skills.

For the local server, Node.js with npx and a running Inngest Dev Server; for the cloud server, an Inngest account (the free plan is enough to get going). Plus an MCP client such as Claude Code, Codex or Cursor.

Inngest offers two official MCP servers: a local one in the Dev Server and a hosted one for Inngest Cloud. This guide shows how to set up both and use them safely.

Step 1: Prerequisites

You need an MCP client such as Claude Code, Codex or Cursor. For the local server, Node.js with npx and a project with Inngest functions are enough. For the cloud server you also need an Inngest account. According to the pricing page, the free plan with no credit card is enough to get started.

Step 2: Start the local server in the Dev Server

Start the Dev Server with npx --ignore-scripts=false inngest-cli@latest dev. The MCP endpoint is then available at http://127.0.0.1:8288/mcp. In Claude Code, register it with claude mcp add --transport http inngest-dev http://127.0.0.1:8288/mcp. No sign-in is needed. If you changed the Dev Server port, adjust the address accordingly.

npx --ignore-scripts=false inngest-cli@latest dev
claude mcp add --transport http inngest-dev http://127.0.0.1:8288/mcp

Step 3: Connect the cloud server in Claude Code

Run claude mcp add --transport http inngest-cloud https://api.inngest.com/mcp. Then call /mcp in Claude Code and authenticate inngest-cloud. In the browser you pick the account, environment, permissions and session length. The official plugin makes it even easier: claude plugin install inngest@anthropic-plugin-directory bundles the cloud server with matching skills.

claude mcp add --transport http inngest-cloud https://api.inngest.com/mcp
claude plugin install inngest@anthropic-plugin-directory

Step 4: Codex and Cursor

In Codex, the command is codex mcp add inngest-cloud --url https://api.inngest.com/mcp, followed by codex mcp login inngest-cloud to sign in. In Cursor, add an inngest-cloud entry with the cloud address to .cursor/mcp.json and, as the documentation describes, set inngest-mcp-static as the client ID. Then reload Cursor and sign in.

codex mcp add inngest-cloud --url https://api.inngest.com/mcp
codex mcp login inngest-cloud

Choose permissions deliberately

The documentation does not mention a dedicated read-only mode. Many tools send events, invoke functions or cancel runs. So in the OAuth session, grant only the environment and permissions you really need, ideally a staging or development environment to begin with. If you use an API key instead of OAuth, the assistant gets broader rights, up to reading signing keys.

Step 5: First real test

Ask the assistant something like: Show me the failed runs from the last hour and explain which step they fail in. It lists the runs, reads the trace tree and links it to your code. After a fix, have it restart the run from the failed step with the rerun tool. Expensive earlier steps then do not run again.

Cloud: OAuth in the browser (recommended, with a choice of account, environment, permissions and session length) or an API key as Authorization: Bearer; signing keys are not accepted. Dev Server: no sign-in.

Read apps, functions, events, runs and traces; send events, invoke functions, cancel and rerun runs. Cloud only: sync apps, manage environments and webhooks, Insights SQL, sessions, experiments and scores, and with an API key also read event and signing keys.

The cloud server is run by Inngest; the Dev Server endpoint only listens locally on 127.0.0.1. Event data, step output and query results the assistant fetches enter the AI client context and therefore reach the model provider.

No read-only mode documented: the assistant can send events, invoke functions, cancel runs and sync apps in production. With an API key it can read event and signing keys. Event payloads may contain personal data.