AWS Builder Center
How to plan re:Invent 2026 with the new AWS Events API and MCP server

How to plan re:Invent 2026 with the new AWS Events API and MCP server

There's a new AWS Events API and MCP server, which means your agent can dig through the whole re:Invent catalog for you. Plus - I made a skill so you don't have to!

Written by Brooke Jamieson, Senior Developer Advocate - AWS and 'voice of developers' for the AWS re:Invent Leadership Team. Last updated 23 September 2026. Everything technical here is from the  AWS Events API Developer Guide  . 
The best part of re:Invent is that there's so many sessions to choose from. But, the worst part of re:Invent is that there's SO MANY SESSIONS TO CHOOSE FROM! It can be kind of overwhelming and hard to sift through if you don't really know what you're looking for, or you don't know how to look. This has (admittedly) been kind of annoying for a while, and so I'm excited to say that there's now an  AWS Events API and MCP server  so your agent can look at the catalog for you and help you plan your week.
In this post, I'll tell you a little more about it, and give you some ideas for how to use it, plus I share my skill that I made for my agents to be able to work with this a little easier.

What is the AWS Events API?

The  AWS Events API  is a REST API and an MCP server for reading the session catalogs of AWS events (re:Invent, Summits, etc) and managing your own schedule, meaning reservations, favourites and personal time. It lives at  api.awsevents.com  and there's also  docs .
The REST API is for scripts and apps.  ListEvents  doesn't need you to sign in, and there's an  OpenAPI spec  if you want to build something of your own.
The MCP server is what most people will probably use, because it means you can give your agent the whole catalog and describe what you want in normal language. Something like "I write Python, I'm interested in agents, I don't want anything before 9am, and I'd like one session that has nothing to do with my job" is a super reasonable request, and the agent can go and work through the catalog and come back with a list.
One thing to know is that the API itself doesn't search or filter. There's no query parameter. It gives you the catalog a page at a time and the filtering happens on your side, which for an agent is fine because that's the part it's good at. If you're writing your own client, the only real gotcha is that pages can come back short and that doesn't mean you're finished. You keep going until there's no  nextToken .

How do I connect the AWS Events MCP server to Kiro?

Add this to your  Kiro MCP config  and reload the window.
{
"mcpServers": {
"awsevents": {
"url": "https://api.awsevents.com/mcp",
"oauth": {
"clientId": "7vmom55m1qstvq8i71ph127bfq",
"redirectUri": "http://127.0.0.1:8976",
"oauthScopes": ["openid", "email", "events/access"]
}
}
}
}
The  redirectUri  needs to end at the port with no path after it. Kiro adds  /oauth/callback  on its own, and if you include a path Kiro can't read the port, picks a random one, and sign-in fails with  redirect_mismatch .
For Claude Code it's one command:
claude mcp add --transport http --scope user awsevents \
https://api.awsevents.com/mcp \
--callback-port 8484 --client-id 7vmom55m1qstvq8i71ph127bfq
The client ID is shared by everyone and is meant to be public. The flow uses PKCE so there's no secret involved. The docs only walk through Kiro and Claude Code, but it's standard MCP over streamable HTTP with OAuth, so Codex, Claude Desktop and other clients that support remote MCP servers should work too.
Then ask your agent to list events. A browser will open, you sign in with your AWS Builder ID, and you're good to go.
A couple of things to be aware of: the MCP server needs you to sign in for every call (even listing events) which is different from the REST API. And you need to already be registered for re:Invent to read its catalog, because the API can't register you, and it won't even give you a link. If you're signed in but not registered you'll get a  403 , and signing in again won't change that, so register on the  re:Invent site  first.

What can I ask it to do?

Anything the catalog website does, plus a lot it doesn't. Some ideas:
  • "Go through the whole re:Invent 2026 catalog and shortlist every 300 or 400 level session about agents or MCP. Skip the abstracts on the first pass and then pull them for the shortlist." There's an  includeAbstracts=false  flag for this, and it makes the first pass a lot quicker because abstracts are the biggest field.
  • "What workshops are on Tuesday afternoon?"
  • "Pick one session on a topic I know nothing about."
  • "Favourite all of those." (It does them ten at a time.)
  • "Show me my schedule and tell me where I've double booked myself."
  • "Block out Wednesday 12 to 1 and call it lunch."
Each session comes back with a title, abstract, code, type, level, tracks, topics, roles, services, start and end times, room, venue and speaker names. So "every session where someone from the Kiro team is speaking" is also a normal thing to ask now.
If you also have the  AWS MCP Server  from the  Agent Toolkit for AWS  connected, your agent can go one step further and look at what you're actually running before it looks at the catalog. So instead of you describing your stack from memory, it can see that you've got Lambda and DynamoDB and a lot of CloudWatch alarms, and go and find the sessions about that. You pick the AWS profile it uses, and the toolkit has IAM condition keys so you can keep the agent read-only even if your own role can write, which is what I'd do for this.

When does re:Invent 2026 reserved seating open?

Reserved seating for re:Invent 2026 opens on 6 October 2026 on the  re:Invent site,  and on 8 October 2026 through the API. Until then, reserving or cancelling through the API returns a  409  and nothing happens. Reading the catalog and favouriting sessions both work right now.
That's why I wanted to get this out in September. If you favourite things now, you'll have your list ready when seating opens, and the popular workshops do fill up quickly.
When you do start reserving, a  200  doesn't mean every session went through. Each one in a batch succeeds or fails on its own (it might be full, clash with something you already have, or already be on your schedule). So when your agent says it's done, ask it to check  GetSchedule  to confirm.

Is there a skill to help agents use the AWS Events API?

I made one so you don't have to. The docs are good, but agents don't always read all of them before they start, and there are a few things that are easy to get wrong: the redirect port, the pagination rule, the  409  before 8 October, the per-session results, and what to do when a write times out. The skill puts all of that in front of the agent up front and tells it to check the live docs for anything that's changed.
Drop it in  ~/.kiro/skills/aws-events-api-mcp/SKILL.md , or wherever your agent keeps its skills. I know I should put this on GitHub. I haven't. Here it is instead.
---
name: aws-events-api-mcp
description: Work with the AWS Events API and its MCP server to discover AWS conference events and sessions, connect Kiro or another MCP client, authenticate an attendee with AWS Builder ID OAuth and PKCE, inspect a personal schedule, or manage reservations, favorites, and personal time. Use this skill whenever a request mentions api.awsevents.com, AWS Events API, AWS conference or re:Invent session catalogs, the awsevents MCP server, event-session reservations, favorites, or personal event schedules. Use it even when the user only asks to find or plan conference sessions, because catalog-access and schedule rules depend on event registration and the chosen REST or MCP interface.
---

# AWS Events API and MCP

Use the canonical AWS Events documentation as the source of truth for this workflow. This API acts only for the signed-in attendee: it can read event catalogs and alter that attendee's own schedule, but it cannot register a person for an event or access another attendee's data.

## Source documentation and freshness

This public skill bundles no AWS documentation. Before answering behavior-specific questions, consult the current [AWS Events Developer Guide](https://docs.aws.amazon.com/events/latest/devguide/what-is-events-api.html) and, for REST contracts, the [live OpenAPI definition](https://api.awsevents.com/v1/openapi.json). Treat event dates, feature availability, operation status, quotas, endpoints, and error behavior as potentially changeable.

If the optional AWS MCP Server from the AWS Agent Toolkit is configured, use its documentation-search and retrieval capability for the current AWS sources. It does not require AWS credentials for documentation search. Do not require that MCP server or confuse it with the separate `awsevents` MCP server: local skills, the AWS documentation MCP, and the AWS Events MCP operate independently. If the AWS MCP Server is unavailable, use the canonical URLs directly and state that live verification could not be performed.

## Choose the interface

Use the MCP server for an AI assistant or agent. It exposes the REST operations as tools, but **every MCP tool call requires sign-in**, including `ListEvents` and catalog reads.

Use REST for scripts or applications. `ListEvents` and `GetEvent` are public. `ListSessions` and `GetSession` are public only for events that do not require registration. All schedule operations require an access token and event registration.

Before choosing an interface when access behavior matters, consult the live [overview](https://docs.aws.amazon.com/events/latest/devguide/what-is-events-api.html) and [MCP-server guide](https://docs.aws.amazon.com/events/latest/devguide/mcp-server.html).

## Connect Kiro to the MCP server

For a request to configure Kiro, add this server entry to the user's MCP configuration. Preserve other server entries. The `redirectUri` must end at the port, without a path; Kiro appends `/oauth/callback` itself. Reload the Kiro window after editing.

```json
{
"mcpServers": {
"awsevents": {
"url": "https://api.awsevents.com/mcp",
"oauth": {
"clientId": "7vmom55m1qstvq8i71ph127bfq",
"redirectUri": "http://127.0.0.1:8976",
"oauthScopes": ["openid", "email", "events/access"]
}
}
}
}
```

Do not choose a random port or add a path to `redirectUri`; either causes a `redirect_mismatch` during sign-in. On first tool use, the MCP client opens the browser sign-in flow and keeps its token locally. Verify connection by calling `ListEvents`.

For other MCP clients, use streamable HTTP endpoint `https://api.awsevents.com/mcp`, shared public client ID `7vmom55m1qstvq8i71ph127bfq`, and a callback URI exactly as registered for that client. The client ID is intentionally public; this OAuth flow uses PKCE and has no client secret. Consult the live [MCP-server guide](https://docs.aws.amazon.com/events/latest/devguide/mcp-server.html) for client-specific setup.

## REST API quick reference

Base URL: `https://api.awsevents.com`; all paths begin `/v1`; requests and responses are JSON. The live OpenAPI description is `https://api.awsevents.com/v1/openapi.json`.

| Intent | REST operation | Authentication |
|---|---|---|
| List current/upcoming events | `GET /v1/events` (`includePast=true` optional) | None |
| Inspect one event | `GET /v1/events/{eventId}` | None |
| List catalog pages | `GET /v1/events/{eventId}/sessions` | Required only when the event requires registration |
| Inspect one session | `GET /v1/events/{eventId}/sessions/{sessionId}` | Required only when the event requires registration |
| Read the attendee's schedule | `GET /v1/events/{eventId}/schedule` | Valid attendee token and registration |
| Reserve 1-10 sessions | `POST /v1/events/{eventId}/reservations` | Valid attendee token and registration |
| Cancel a reservation | `DELETE /v1/events/{eventId}/reservations/{sessionId}` | Valid attendee token and registration |
| Favorite 1-10 sessions | `POST /v1/events/{eventId}/favorites` | Valid attendee token and registration |
| Remove a favorite | `DELETE /v1/events/{eventId}/favorites/{sessionId}` | Valid attendee token and registration |
| Add personal time | `POST /v1/events/{eventId}/personal-time` | Valid attendee token and registration |
| Replace personal time | `PUT /v1/events/{eventId}/personal-time/{personalTimeId}` | Valid attendee token and registration |
| Delete personal time | `DELETE /v1/events/{eventId}/personal-time/{personalTimeId}` | Valid attendee token and registration |

For exact query parameters, field shapes, and examples, consult the live [REST API guide](https://docs.aws.amazon.com/events/latest/devguide/rest-api.html) and [OpenAPI definition](https://api.awsevents.com/v1/openapi.json).

* Traverse every `ListSessions` page until `nextToken` is **absent**. A short page is not the last page.
* Use `includeAbstracts=false` when abstracts are unnecessary, and `locale` for a requested language. Treat session fields as optional; event catalogs can omit fields.
* Treat `GetSchedule` as the source of truth. Use it to confirm schedule changes and obtain IDs omitted by write responses.

## Access and authentication

Before an authenticated REST call, consult the live [authentication guide](https://docs.aws.amazon.com/events/latest/devguide/authentication.html) and its linked details.

1. Determine whether the event requires registration from `ListEvents` or `GetEvent`.
2. For protected catalog reads and all schedule operations, the attendee needs both a valid access token and registration for the selected event. A valid token without registration gives `403`; signing in again does not fix it.
3. Send only the access token in `Authorization: Bearer $ACCESS_TOKEN` to `https://api.awsevents.com`.
4. For an application implementing sign-in, use OAuth authorization-code flow with PKCE, a fresh cryptographic verifier and state for every attempt, S256 base64url challenge without padding, `identity_provider=AWSBuilderID`, and a local exact-match redirect URI. Application callback ports are `8484` through `8489`, with `/callback`.
5. Refresh with the refresh token based on `expires_in`. On a `401`, refresh once and retry once. If it still returns `401`, require interactive sign-in rather than looping.

Access tokens last 60 minutes; refresh tokens last 30 days. Never log tokens, put them in URLs, or expose them in generated code, reports, or chat output. Browser apps keep tokens only in memory; other local apps use the OS keychain or a user-readable-only file. Consult the live [token-handling guidance](https://docs.aws.amazon.com/events/latest/devguide/auth-handling-tokens.html) for storage rules.

To sign out fully, revoke the refresh token, discard local tokens, and end the broker and Builder ID browser sessions. Consult the live [sign-out guidance](https://docs.aws.amazon.com/events/latest/devguide/auth-signing-out.html) before implementing this flow. Deleting an MCP configuration does not revoke its stored token.

## Schedule-changing operations

Reservation, favorite, and personal-time changes have attendee-visible effects. Confirm the intended event and exact sessions or personal-time details before a change unless the user already stated them unambiguously. If a reservation or favorite request lacks an unambiguous `eventId`, exact `sessionIds`, or required confirmation, state that no change occurred and request those missing inputs. For a later execution, state the full completion sequence: submit the change, inspect each per-session result, then use `GetSchedule` before reporting the final outcome. For a bulk reservation/favorite response, inspect each per-session success or failure: HTTP `200` does not mean every item succeeded. Confirm final state with `GetSchedule` and report successes, failures, and relevant clash information.

Use these request constraints:

* `ReserveSessions` and `AssociateFavorites`: send 1-10 distinct `sessionIds`; a favorite records interest and does not reserve a seat.
* `CancelReservation` and `DisassociateFavorite`: a `404` means the item is already absent and can be treated as complete on retry.
* `CreatePersonalTime`: `title` is 1-128 characters, `description` 1-250; timestamps are UTC `YYYY-MM-DDTHH:mm:ss` without `Z` or offset, seconds must be `00`, end must be after start, and duration must be a whole 5-minute increment. `location` is optional and at most 255 characters. Create returns no body, so retrieve the new ID from `GetSchedule`.
* `UpdatePersonalTime` replaces rather than merges. Send every field; omitting `location` clears it.

The service has no idempotency key. After an unknown write outcome (timeout, dropped connection, `500`, or `503`), do not repeat the original request. First read `GetSchedule`, compare it with the intended state, and submit only what remains. Only `CancelReservation` and `DisassociateFavorite` are direct-retry exceptions: retry later and treat `404` as already complete. For `DeletePersonalTime`, reconcile with `GetSchedule` before deciding whether another request is needed.

## Errors, quotas, and reporting

Consult the live [error guide](https://docs.aws.amazon.com/events/latest/devguide/errors.html) and [quota guide](https://docs.aws.amazon.com/events/latest/devguide/quotas.html) for troubleshooting.

* `400`: correct the malformed request; do not retry.
* `401`: sign in or refresh once, then retry once.
* `403` with JSON: attendee is not registered; do not retry. `403` without a body: edge refusal; slow down and retry.
* `404`: not found; treat a removal as already complete.
* `409`: operation temporarily closed; retry only after it reopens. It is not a malformed request.
* `429`: honor `Retry-After` in seconds. Batch writes count every named session toward quota, so batching saves round trips but not quota.
* `500`/`503`: back off reads; reconcile writes with `GetSchedule` before any follow-up change.

When presenting results, state the event ID, interface used, sign-in/registration constraint if applicable, operation outcome, per-session failures for bulk operations, and whether `GetSchedule` confirmed a write. Do not include token values.

## Canonical source map

* Service scope and access rules: https://docs.aws.amazon.com/events/latest/devguide/what-is-events-api.html
* Getting started: https://docs.aws.amazon.com/events/latest/devguide/getting-started.html
* MCP connection and tool list: https://docs.aws.amazon.com/events/latest/devguide/mcp-server.html
* OAuth and token handling: https://docs.aws.amazon.com/events/latest/devguide/authentication.html
* REST contracts: https://docs.aws.amazon.com/events/latest/devguide/rest-api.html and https://api.awsevents.com/v1/openapi.json
* Quotas and error recovery: https://docs.aws.amazon.com/events/latest/devguide/quotas.html and https://docs.aws.amazon.com/events/latest/devguide/errors.html

No AWS documentation is redistributed in this package. Consult the canonical sources above before redistributing AWS documentation or other AWS Content.


A lot of people at AWS have spent the whole year working on what's going to be in those rooms, and I'd like more of it to get found than a search box usually allows for! So connect it, have a look through the catalog with your agent, and favourite a few things before October 6.



About the Author:

Brooke Jamieson is a Senior Developer Advocate at AWS, focused on AI agents and developer tools. A former fashion model turned mathematician turned AI Engineer, Brooke moved from Australia to New York City for this role. They make byte-sized tech content about AI and AWS, and you can find them on  LinkedIn ,  Instagram  or  X .
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article