---
title: "Cloud Agent tasks, a better `railway ca`, Agent Connectors"
date: 2026-11-13T19:16:00.000Z
number: 0316
url: https://railway.com/changelog/2026-11-13-cloud-agent-tasks
---

# Cloud Agent tasks, a better `railway ca`, Agent Connectors

Give your agents the work, then get back to yours. This week, [Tasks](https://docs.railway.com/cloud-agents/tasks) bring cloud agents into your scripts and external workflows, [`railway ca`](https://docs.railway.com/cli/ca) puts the threads that need you first, and [Agent Connectors](https://docs.railway.com/ai/agent-connectors) become generally available.

# Cloud Agent tasks

[Hand a task to a Railway cloud agent](https://cms.railway.com/media/324a49f6-ebe7-401a-8fe0-1352bf76d4f6-cloud-agent-tasks-purple-4k.mp4)

Hand a coding task to a [Railway cloud agent](https://docs.railway.com/cloud-agents) through [MCP](https://docs.railway.com/ai/mcp-server), the [CLI](https://docs.railway.com/cli/ca#tasks), or the [public API](https://docs.railway.com/integrations/api/manage-cloud-agent-tasks). Your editor, a script, or another system can send the work and pick up the result without opening a remote session.

Tasks run in the background, keep their results, and support follow-ups in the same conversation. Check progress, cancel a run, or receive [webhook updates](https://docs.railway.com/integrations/api/manage-cloud-agent-tasks) from your integration. Add `--interactive` to let the agent pause for a question.

From your terminal, start with `railway ca task run "Fix the failing test in api/ and open a pull request" --repo acme/api`. The command returns a task ID. Use `railway ca task status <task-id>` to read the result, or add `--watch` to follow the work live.

Starting a fresh agent creates a billed VM. In scripts, pass `--accept` to approve creating one. To continue earlier work with its context, reply to a task or reuse its conversation key.

Read the [Tasks guide](https://docs.railway.com/cloud-agents/tasks), and share what you're building on [Central Station](https://station.railway.com/new?type=feedback).

# A better `railway ca`

[Browse prioritized threads and check cloud agent status from your terminal](https://cms.railway.com/media/094534e9-ca6b-4d5e-8cac-dc697ef9ccfc-railway-ca-tui-redo-4k.mp4)

`railway ca` puts conversations ahead of machines, with threads grouped by what needs you, what's working, and what's done. Machines and projects sit in collapsible groups below them.

Automation runs appear in the same sidebar. Your existing harness options stay available, with `railway agent` as the default. Saved preferences still apply.

Each thread shows its work status, so you can spot blocked work or errors without opening every session. Machine status also shows whether an agent is awake, sleeping, waking, or down.

Press Enter on a sleeping thread to wake its machine and reopen the conversation. An opening card shows progress as Railway reconnects. You can also archive agents you're finished with and unarchive them when you need them again.

For a quick check outside the interface, `railway ca list` includes a `STATE` column.

Run `railway upgrade`, then open `railway ca`. Tell us how it works for your workflow on [Central Station](https://station.railway.com/new?type=feedback).

# Agent Connectors

[Choose connectors for a Railway Agent conversation](https://cms.railway.com/media/agent-connectors.mp4)

Agent Connectors are now generally available. You no longer need Priority Boarding or a feature flag to connect external tools to Railway Agent.

Add a connection in [your account settings](https://railway.com/account/agents), then select it from **Contexts** in a Railway Agent chat.

See [supported connectors and setup](https://docs.railway.com/ai/agent-connectors), and share your feedback on [Central Station](https://station.railway.com/new?type=feedback).

# Fixes and improvements

- We fixed service variable search accepting text without updating its results
- We fixed discarded [staged changes](https://docs.railway.com/deployments/staged-changes) briefly reappearing when an older update arrived
- We fixed the missing **Leave Workspace** action for members with the Deployer role