# AI and Asana.

URL: https://levystreet.com/integrations/asana/
Updated: 2026-10-19

An AI Asana integration guide for NZ teams: AI Teammates and AI request pricing, the API, webhooks and MCP server, and turning meeting notes into tasks.

## Short answer

An AI Asana integration lets a model create and update tasks from meeting notes, emails or forms, either through Asana's AI Teammates and AI Studio or through the API, webhooks and Asana's MCP server. For New Zealand teams the reliable pattern is extraction with evidence, owner and date matching in code, and a person approving the task list before anything is created.

- Paid Asana plans include five AI requests per user monthly, capped at 50 per account.
- Extra AI requests cost $0.50 prepaid or $0.60 pay-as-you-go; the currency is unstated.
- Asana's MCP V2 server is at mcp.asana.com/v2/mcp; the beta SSE server is deprecated.
- Asana webhooks are at-most-once, so events can be lost; add a polling fallback.
- The API allows 1,500 requests a minute on paid plans, but search only 60.

## What Asana AI does natively (October 2026)

Asana sells three AI layers. AI Teammates are 30+ prebuilt agents with skills such as brief writing and meeting summary and action capture. Asana Dash summarises meetings, emails and tasks. AI Studio builds no-code AI workflows on a separate credit pool. On [Asana's pricing page](https://asana.com/pricing), shown to NZ visitors in Australian dollars on 19 October 2026, Starter is A$12.99 and Advanced A$27.49 per user per month billed annually.

Each paid plan includes five AI Teammates and Dash requests per user per month, capped at 50 per account (250 on Enterprise), plus 50K to 200K AI Studio credits per billing account. Extra requests cost $0.50 prepaid or $0.60 pay-as-you-go. A decision rule: 50 requests is enough to trial meeting capture with one team. If a 30-person firm runs 80 captured meetings a month and each took one request (our assumption), most would be paid overage.

## Integration routes and when to use each

| Route | What it gives you | Use it when |
|---|---|---|
| AI Teammates, Dash and AI Studio | Prebuilt agents and no-code AI rules inside Asana | Usage fits the allowance and setup time matters most |
| REST API | Tasks, projects, sections, users and custom fields through a token or OAuth | You want your own extraction, matching and approval |
| Webhooks | Compact change events after an `X-Hook-Secret` handshake, signed with `X-Hook-Signature` | You react to task changes, with polling as a backstop |
| MCP V2 server | Task and project tools for Claude, ChatGPT and other MCP clients | Individuals capture actions from their own assistant |
| n8n, Make or Zapier | Prebuilt Asana steps | Simple form-to-task or email-to-task flows |

## Automation 1: meeting notes to approved tasks


- **Inputs**: The transcript or notes, meeting date, attendee emails and the target project.
- **Model step**: Extract agreed actions only, as JSON: title, owner as said, due date as said, a verbatim source quote and a confidence score. Discussion that was not agreed is ignored.
- **Matching in code**: Owners resolve to Asana users by attendee email. An ambiguous "Sam" stays unassigned. "Next Friday" is resolved from the meeting date in Pacific/Auckland time.
- **Human approval**: The organiser gets the list in Teams, Slack or email and ticks, edits or deletes each item.
- **Write-back**: Approved items are created in a "From meetings" section, with the quote and a link to the notes in the description.
- **Failure and recovery**: A "Source ref" custom field holds meeting ID plus item number, so a retried run updates instead of duplicating. Wrong tasks are deleted by the organiser; the notes stay the record.


| Extracted field | Asana field | Rule |
|---|---|---|
| `title` | Task name | Starts with a verb, under 80 characters |
| `owner_email` | Assignee | Must match an attendee, otherwise blank |
| `due_date` | Due date | Blank when the notes are vague |
| `source_quote` and link | Description | Always included |
| meeting ID and item | Source ref (custom field) | Prevents duplicates |

The same workflow is mapped for consultancies in [meeting actions into project systems](/industries/professional-services/meeting-actions-into-project-systems/), and our lesson on [turning meeting notes into actions](/learn/turning-meeting-notes-into-actions/) covers the manual version.

## Automation 2: a Friday project status draft


1. **Collect.** On Friday morning, read tasks completed this week, overdue tasks and new blockers for each project.
2. **Draft.** The model writes a short status in the project's usual format, citing task links, with counts calculated in code.
3. **Approve.** The project lead edits and posts it as the project's status update; nothing posts on its own.


## Permissions and approval design

A personal access token acts as the person who created it, so create the integration under a dedicated user who belongs only to the projects in scope, or use OAuth with the narrowest scopes. For MCP, the [developer docs](https://developers.asana.com/docs/using-asanas-mcp-server) say Enterprise+ and Legacy Enterprise customers can allow or block each MCP client through app management; other tiers ask Asana Support. Approval before creation is the control that matters most, as in our [AI workflow automation](/services/ai-workflow-automation/) builds.

Transcripts hold personal information about attendees and the people they discuss. Under IPP 3 of the Privacy Act 2020, take reasonable steps so attendees know their meeting is recorded and processed by AI, and keep the transcript only as long as you need it.

## Costs and limits (as listed 19 October 2026)

- **API**: 1,500 requests per minute per token on paid domains and 150 on free; search is limited to 60 per minute; at most 50 concurrent reads and 15 concurrent writes.
- **Model cost, worked example**: 80 meetings a month at 12,000 input and 800 output tokens on Claude Sonnet 5.5 (US$2 and US$10 per million) is about 80 × US$0.032 = US$2.56.

## Where it goes wrong

- **Lost events.** The [webhooks guide](https://developers.asana.com/docs/webhooks-guide) says delivery is designed as at-most-once, and an unanswered heartbeat for 24 hours deletes the webhook. Reconcile with a scheduled poll.
- **Over-extraction.** "We should look at pricing" becomes a task. Require an explicit commitment and a confidence threshold.
- **Search limits.** Matching owners or projects through search hits the 60-per-minute cap first; cache users and projects instead.
- **Old MCP URLs.** Clients still on the deprecated SSE server lose access when it shuts down.

Teams in Microsoft 365 usually send the approval card through our [Microsoft Teams](/integrations/microsoft-teams/) pattern; teams on boards rather than lists should read the [monday.com guide](/integrations/monday/).

## Questions people ask

### Can Asana's own AI turn meeting notes into tasks?

Asana lists meeting summary and action capture among the skills of its prebuilt AI Teammates, and Asana Dash summarises meetings, emails and tasks. The constraint is volume: paid plans include five AI requests per user per month, capped at 50 per account (250 on Enterprise). For one team trialling capture that may be enough; for a whole firm, budget for extra requests or a custom route.

### How much do Asana AI requests cost in New Zealand?

As listed on 19 October 2026, Asana shows NZ visitors prices in Australian dollars: Starter A$12.99 and Advanced A$27.49 per user per month billed annually. Extra AI requests are $0.50 each prepaid or $0.60 pay-as-you-go, shown as '$' without a currency code. Confirm the currency on your quote, and note that AI Studio uses a separate monthly credit pool.

### Can Claude or ChatGPT create Asana tasks?

Yes. Asana's MCP V2 server at https://mcp.asana.com/v2/mcp lets compatible clients search, create and manage tasks and projects after the user authorises access, and Asana offers packaged apps in Claude and ChatGPT. On Enterprise+ and Legacy Enterprise, admins can allow or block individual MCP clients through app management; other tiers need a support request.

### How do we stop AI creating duplicate or wrong tasks?

Require approval before creation, store a source reference for each item (meeting ID plus item number) so retries update rather than duplicate, and only extract commitments that someone actually agreed to. Leave the owner blank when a name is ambiguous instead of guessing, and resolve relative dates like 'next Friday' from the meeting date in New Zealand time.


## Sources

- [Asana pricing](https://asana.com/pricing)
- [AI Teammates](https://asana.com/product/ai/ai-teammates)
- [Rate limits](https://developers.asana.com/docs/rate-limits)
- [Using Asana's MCP server](https://developers.asana.com/docs/using-asanas-mcp-server)
- [Webhooks guide](https://developers.asana.com/docs/webhooks-guide)
- [Privacy Act 2020](https://www.legislation.govt.nz/act/public/2020/31/en/latest/)
- [Pricing](https://platform.claude.com/docs/en/about-claude/pricing)

Published by Levy Street (https://levystreet.com), Wellington, New Zealand.