Cursor Automations in v3.5: How to Set Up No-Repo and Multi-Repo Agents (May 2026)

Cursor 3.5 shipped on May 20, 2026 with three changes to Automations. Automations moved into the Agents Window inside the IDE. Multi-repo mode arrived, letting a single Automation reason across multiple codebases. And no-repo mode launched: a Cursor Automation can now run with zero codebase attached, monitoring Slack channels and Stripe data and acting on what it finds.

That third mode is the one worth understanding first. Five ready-to-use templates are in the Cursor Marketplace right now. Two of them, the Slack digest agent and the product finance agent, cover workflows most engineering teams currently handle in Zapier or a spreadsheet. This guide walks through setting up both, then covers multi-repo configuration for teams running coordinated changes across repos.

What changed in Cursor 3.5 for Automations

Automations now live in the Agents Window

Before 3.5, managing Automations meant opening a browser tab at cursor.com/automations. That still works, but the Agents Window in the IDE now has an Automations tab. Create, edit, monitor, and activate Automations without leaving the editor. For a team already living inside Cursor all day, this reduces the context switch from "agent ran something" to "I want to check or change what it ran."

No-repo mode: run agents without a codebase

Every Automation before 3.5 required at least one attached repository. The agent needed source code as anchor. No-repo mode removes that requirement entirely. An Automation in this mode connects directly to external tools, Slack, Stripe, Databricks, and acts on the signals those tools surface. No code. No repo attachment.

This is a different product category than what Cursor Automations was six months ago. A Slack digest agent that summarizes your morning messages does not need to read your codebase. A Stripe revenue report does not either. No-repo mode puts Cursor Automations in direct competition with Zapier and n8n for the first time on pure workflow automation tasks.

Multi-repo mode: one agent across multiple codebases

On the other end, multi-repo mode lets a single Automation attach to several repositories at once. The agent sees the file trees and symbol graphs for every attached repo. It can open PRs in two repos, run tests across both, and verify the result before marking the task complete.

The concrete use case: an Automation watches for Linear issues flagging API contract mismatches, then opens coordinated PRs in the frontend and backend repos to fix the mismatch. One Automation, two repos, one verified delivery. That task used to require a human to context-switch between repos and coordinate both sides of the change.

We covered the environment configuration that multi-repo Automations build on in our Cursor 3.4 cloud agent environments guide. The Dockerfile-based environment setup from 3.4 is what gives multi-repo agents a clean, reproducible execution context.

Five Marketplace templates and the seven-day discount

The Cursor Marketplace now lists five no-repo Automation templates, per the official changelog:

• Slack digest agent - summarizes unread messages and channels by priority each morning
• Product analytics agent - weekly metrics digest from a data warehouse like Databricks
• Product FAQ agent - watches a Slack channel for questions and drafts answers from docs and past threads
• Product finance agent - pulls Stripe billing data for recurring revenue reports
• Customer health monitoring agent - tracks signals across Granola, Slack, and Databricks and flags at-risk accounts

All agent runs on newly created Automations are 50% off for seven days from the May 20 release date. The discount window closes May 27. More on this in the FAQ.

Setting up the Slack digest agent (no-repo template)

The Slack digest agent is a practical starting point for seeing what no-repo mode actually does. The setup process is short, assuming your Slack workspace admin can approve the OAuth request promptly.

Find Automations in the Agents Window

Open Cursor 3.5. In the Agents Window, click the Automations tab. If the tab isn't visible, confirm your version via Cursor > About Cursor. The Automations tab was not present before 3.5.

Click New Automation.

Choose the Slack digest template

In the template picker, click Browse Marketplace. Filter by No-Repo to see only templates that run without a codebase attached. Select Slack Digest Agent.

Connect your Slack workspace

Click Connect Slack and complete the OAuth flow. The agent requests these Slack permission scopes during authorization:

• channels:history - read message history in public channels
• channels:read - list public channels
• chat:write - post the digest summary to an output channel
• groups:read - private channels (optional; skip if you only need public channels)
• im:history - direct messages (optional; only needed for DM summary)

Minimum required for a public-channel-only digest: channels:history, channels:read, and chat:write. Your Slack workspace admin will see these during the authorization flow and can approve or restrict from there.

Set the schedule and activate

The schedule field accepts cron syntax. For a 9am weekday digest in Eastern time:

{
  "schedule": "0 9 * * 1-5",
  "timezone": "America/New_York",
  "channels": ["#general", "#engineering", "#product"],
  "output_channel": "#morning-digest",
  "summary_format": "priority_ranked"
}

The priority_ranked format orders channel summaries by a combination of mention count, reaction count, and thread depth. High-engagement threads appear first; low-activity channels are deprioritized automatically.

Click Activate. The first run triggers at the next scheduled time. Click Run Now to test before then.

What the agent outputs and how to tune channel weights

The digest posts to your output channel as a structured Slack message. Each included channel gets a summary block with the top 3-5 threads, sender names, message previews, and links back to the original threads.

To tune which channels get more weight in the summary, add a channel_weights map:

{
  "channels": ["#general", "#engineering", "#incidents"],
  "channel_weights": {
    "#incidents": 2.0,
    "#general": 0.5
  }
}

With #incidents weighted at 2.0, the digest surfaces incident threads even when the raw activity count is lower than a busier but lower-priority channel.

Activating the product finance agent (Stripe template)

The product finance agent connects to Stripe and delivers a scheduled revenue report. Setup follows the same flow as the Slack digest agent, with one addition: you'll create a restricted Stripe API key before starting.

Connect a Stripe account

In the template configuration, click Connect Stripe. The agent uses a restricted key, not your main secret key. Create one in the Stripe Dashboard under Developers > API keys > Restricted keys with these permissions:

Read: Charges
Read: Customers
Read: Invoices
Read: Subscriptions

Paste the restricted key into the credentials field. Cursor stores it as an encrypted environment variable scoped to this Automation. The key is not visible in your Automation config after saving.

Configure the report scope

The default template reports on monthly recurring revenue (MRR), active subscription count, and recent churn. Extend the scope in the configuration:

{
  "report_scope": ["mrr", "active_subscriptions", "churn", "expansion_revenue"],
  "schedule": "0 8 * * 1",
  "output_channel": "#finance-digest",
  "currency": "USD",
  "lookback_days": 7
}

Adding expansion_revenue captures upgrades from one plan tier to another, which the default template skips. For a SaaS team tracking net revenue retention, that addition makes the weekly report meaningfully more complete.

How no-repo mode works here

The product finance agent shows what no-repo mode means in practice. Per the Cursor 3.5 changelog, the product finance template pulls billing data from Stripe for revenue reporting, with no codebase attachment required. The revenue data lives in Stripe regardless of what language or framework your billing integration uses.

The scope of what this Automation does is read-only. It surfaces Stripe data for team visibility. It does not write to Stripe or change any billing configuration.

Multi-repo Automations for engineering teams

Multi-repo mode is available on the Teams plan ($40/user/month), which includes "cloud agents with shared team context" and "team-wide rules, skills, and automations." If you're on the Individual plan and want multi-repo, check the current pricing page for what's included at your tier.

Attach a second repo to an existing Automation

Open any Automation in the Agents Window. Under Configuration > Repositories, click Add Repository. You can attach any repo your Cursor team account has access to. Each repo can be pinned to a specific branch or set to track the default branch.

{
  "repos": [
    {
      "url": "https://github.com/your-org/frontend",
      "branch": "main"
    },
    {
      "url": "https://github.com/your-org/api-service",
      "branch": "main"
    }
  ]
}

After saving, the Automation's context includes the full file tree, symbol graph, and recent commit history for both repos.

What "reason across all required context" means

The Cursor changelog describes multi-repo agents as able to "reason across all required context and work across repos to deliver, test, and verify tasks." In the API contract mismatch example: the agent reads the type definition in api-service, traces how the frontend calls that endpoint in frontend, writes the changes in both repos, opens two PRs, and runs the test suite in each before posting a summary.

The agent does not merge PRs autonomously. It opens the PRs, runs the tests, and delivers a status message with links to both PRs. Human review and merge is still the final step. The delivery is automated; the approval is not.

A practical use case: coordinated frontend/API changes

Say a Linear issue arrives tagged api-contract-change. The Automation trigger fires. The agent reads the issue, identifies the affected endpoint in api-service, finds the frontend calls to that endpoint in frontend, opens coordinated PRs in both repos with the required changes, and posts a Slack message with both PR links and the test results.

If your team runs on Jira instead of Linear, the same pattern works. The Cursor Jira cloud agent guide covers how Jira task context feeds into agent runs, including how issue fields map to agent context.

This is the task that has no Zapier equivalent. Zapier can trigger on a Linear event. It cannot reason about what the API contract change means across two codebases, write the changes, and verify the tests. That requires codebase context.

Cursor Automations vs Zapier vs n8n

The five no-repo templates create genuine overlap with workflows teams already run in Zapier and n8n. The boundary is clearer than the feature names suggest.

When Cursor Automations is the right call

Use Cursor Automations when the task needs codebase context. Bug reports that require reading source code, API changes that span two repos, CI failures that need diagnosis against actual test files. These can't run in Zapier because Zapier has no concept of a codebase.

Even the no-repo templates benefit from being inside Cursor. You can extend the Slack digest agent to include recent commit summaries and open PR status alongside the channel digest. That combination is not possible in Zapier or n8n without building a custom integration.

The Zapier MCP guide we published last week is worth reading alongside this one if your team already has Zapier's MCP integration running with AI tools. The two can complement each other rather than compete.

When Zapier or n8n is still the better choice

Pure trigger-action pipelines with no code context needed. Zapier's strengths are its connector breadth (6,000+ apps) and the reliability of its execution model for simple workflows. If you want a Slack notification when a Stripe payment fails, Zapier does that in three clicks and runs it reliably for years without maintenance.

n8n is the right pick when your team wants self-hosted automation infrastructure with fine-grained data transformation between steps. Its visual workflow builder and the ability to run your own instance with full data control are advantages Cursor Automations doesn't offer. Teams building ETL-adjacent pipelines from Databricks or Postgres into reporting systems will find n8n a better fit there.

When you stack them

A Zapier Zap detects a customer support ticket in Zendesk, fires a webhook, and a Cursor Automation picks it up and reads the customer's codebase context to help draft a technical response. The trigger-action layer is Zapier. The codebase-aware reasoning layer is Cursor. The two have complementary strengths; running them as a stack is not a workaround, it's a reasonable architecture for developer-tools companies.

Task	Cursor Automations	Zapier	n8n
Fix a bug from a Linear issue	Yes - codebase context required	No	No
Multi-repo PR coordination	Yes (Teams plan)	No	No
Slack channel morning digest	Yes (no-repo template)	Yes	Yes
Weekly Stripe MRR report	Yes (no-repo template)	Yes	Yes
Self-hosted data pipeline	No	No	Yes
6,000+ pre-built app connectors	No	Yes	Partial
Webhook trigger into codebase action	Yes	Trigger only	Trigger only

The direct read: Cursor Automations wins on any task that needs to read or change code. Zapier wins on connector breadth and setup simplicity for pure trigger-action pipelines. n8n wins when self-hosted control and data transform complexity are the requirements. The no-repo templates in 3.5 move Cursor into Zapier/n8n territory for a handful of specific use cases; they don't replace either tool for the workflows those tools already handle well.

Frequently asked questions

Is the 50%-off discount still running?

The discount applies to agent runs on newly created Automations for seven days from May 20, 2026. If you're reading this before May 27, yes. After May 27, standard agent run pricing applies. Check cursor.com/pricing for current per-run costs.

Which Cursor plans include Automations?

Cloud agents (which Automations run on) are available on the Individual plan ($20/month) and the Teams plan ($40/user/month), per the current pricing page. The Teams plan adds "team-wide rules, skills, and automations" and "cloud agents with shared team context." Multi-repo shared context appears to be a Teams-tier capability. Free and Hobby plans do not include cloud agents.

What Slack permissions does the Slack digest agent need?

Minimum set for public channels: channels:history, channels:read, chat:write. Add groups:read for private channels and im:history for direct messages. The OAuth authorization screen shows exactly what the agent is requesting before you approve.

Can I build a custom no-repo Automation without using a template?

Yes. In the New Automation flow, select Blank instead of a Marketplace template. You define the trigger (schedule or webhook), the tools the agent can access via OAuth (Slack, Stripe, etc.), and the task prompt. The five Marketplace templates are starting points; the full surface area of no-repo mode is whatever tools you can connect via OAuth.

Does multi-repo Automations require a Teams plan?

Based on the current pricing page, "team-wide rules, skills, and automations" and "cloud agents with shared team context" are Teams plan features. If you're on an Individual plan and want to test multi-repo mode, verify what's available at your tier at cursor.com/pricing before investing setup time.