Cursor 3.5's No-Repo Automations: Agents That Watch, Not Code

On May 20, Cursor 3.5 expanded Cursor Automations in a direction no one was expecting: agents that don't touch code. Until this release, every automation needed a repository — the agent existed to open PRs against your codebase on a schedule. Now you can build automations with no repo attached, designed to watch external systems and surface what matters.

Cursor shipped five no-repo templates in the Marketplace at launch: a Slack digest agent that summarizes channel activity, a product analytics agent that pulls weekly metrics from your warehouse, a product FAQ agent that answers questions from docs, a finance agent that builds revenue reports from billing data, and a customer health agent that flags accounts whose usage signals are shifting.

Why This Matters

The interesting part isn't the templates themselves — it's that Cursor's agent infrastructure now runs as a general-purpose scheduled task runner with LLM reasoning built in. You can connect MCP servers to it, give it a system prompt, set a cron schedule, and have it report results into Slack, email, or a webhook. It's effectively a managed Zapier with reasoning, billed against your existing Cursor usage.

If you've been building little Python scripts on a Raspberry Pi or paying for n8n cloud to do this kind of work, Cursor 3.5 closes that gap for any team already on a paid Cursor plan.

Setting One Up

Open cursor.com/automations or the Agents Window in the editor and click Create. Pick a template or start blank. The configuration form asks for a trigger (schedule, webhook, or manual), one or more MCP servers the agent should have access to, and a system prompt describing the job.

Trigger: Weekly, Monday 9am ET
MCP servers: slack, posthog
System prompt:
  You are a product analytics digest writer. Each Monday,
  pull the past week's PostHog data for product_signup, paid_conversion,
  and weekly_active_users. Compare to the prior week.
  Post a 5-line summary to #product-metrics in Slack.
  Lead with the metric that moved the most.
Output: Slack message to #product-metrics

Multi-Repo and Cross-Repo Work

The same release also added multi-repo support to regular automations. An agent can now attach more than one repository — useful when a feature spans your backend and your client SDK, or when an automation should test against both your main app and your design system. The agent can read, edit, and open PRs in any of the attached repos in a single run.

There's a soft cap of five repos per automation today. The agent picks which ones it actually touches based on the task, so you can attach everything that might be relevant without paying for context you don't use.

Pricing and the 50% Discount

All agent runs for newly created automations are 50% off through May 27. Standard rates run on the Composer 2.5 price ladder — $0.50 per million input tokens, $2.50 per million output. Fast mode (10x throughput) is $3 and $15 respectively. A typical no-repo Slack digest agent that runs daily and reads maybe 50K tokens of context will cost under $1 a month at the standard tier, even after the discount expires.

Where the Templates Get Tricky

The product FAQ agent is the one most teams will reach for and the one most likely to disappoint without tuning. By default it answers from your docs via MCP, but it has no concept of "this question is out of scope" — if a customer asks about pricing in your #support channel, it will guess from whatever doc is closest. Add a clear refusal rule in the system prompt for question types the docs don't actually cover.

The customer health agent is the opposite — it's narrow and good at one thing (flagging accounts with usage drops or support ticket spikes), but it needs a tight MCP connection to your CRM or product analytics. If you don't have that data flowing into an MCP server, the agent has nothing to work with.

The Broader Pattern

Cursor 3.5 quietly shifted the product from "AI editor that can also schedule code work" to "agent platform that happens to include an editor." The no-repo templates are a marketing wedge, but the underlying capability — scheduled MCP-aware agents with team-wide audit logs — is the actual shipped feature. If you've been waiting for a reason to put internal ops agents behind a managed service, this is it.