The MCP for cloud automations.

Save the agent session that worked as a workflow. Kavela runs it in the cloud on your schedule and checks every run.

Claude Code~/work/ai-newskavela connected

That's the format I want. Run this every weekday at 8, same output.

I'll save this session as a Kavela workflow.

kavela - create_workflow (MCP)(name: "Morning AI news TLDR", starts: "schedule", atTimes: ["08:00"], days: "weekdays")

Saved the workflow Morning AI news TLDR. It runs weekdays at 08:00 (Asia/Singapore). Kav picks who does the work on each run. Run it now with run_workflow, or see how it goes with workflow_status.

Add a checklist: the file is written today, exactly 3 posts, each with its link, none older than 24 hours.

kavela - update_workflow (MCP)(workflow: "Morning AI news TLDR", version: 1, checks: [4 lines])

Saved Morning AI news TLDR as version 2. Changed: the checklist. The next run uses it.

Piloting with

Only 6.6% of developers trust AI with important work decisions. Nearly half trust it only when they can check it. Kavela checks every run, so you don't have to.

Stack Overflow Developer Survey 2026

Save the session that worked.

When a session gets the job right, ask your agent to save it. It writes the workflow from what you just did, including what to deliver and what it must never do.

On your schedule: Set times on set days, every few minutes, whenever another service sends data, or only when you run it.
Kav does each run: Kav, the workspace manager, picks the right agents for every run.
Also in Kavela: The workflow, its runs and its files are in your Kavela workspace too.

Morning AI news TLDROn a schedule, weekdays at 08:00

Workflow

Every weekday morning, pick the 3 front-page Hacker News posts most relevant to AI agents, skipping anything older than 24 hours.

I'll save this session as a Kavela workflow.

kavela - create_workflow (MCP)(name: "Morning AI news TLDR", starts: "schedule", atTimes: ["08:00"], days: "weekdays")

Saved the workflow Morning AI news TLDR. It runs weekdays at 08:00 (Asia/Singapore). Kav picks who does the work on each run. Run it now with run_workflow, or see how it goes with workflow_status.

It runs in the cloud. Close your laptop.

Every run happens on Kavela's agents, each with its own computer and a real browser, working in the tools your workspace connected.

On your own plan: Runs on your Claude or Codex subscription, or on the Kav AI usage your plan includes.
Tools connected once: Apps connect to the workspace once, and every agent can use them.
Decisions come to you: Anything that needs you waits in Kavela, and your agent can list it with list_approvals.

Every run is checked. A miss never passes quietly.

A second agent checks each run against your checklist, or against the instruction when there is none. Ask your agent how it's going and it reads you the results.

Passed, failed or didn't finish: workflow_status lists the latest runs and how many passed on each version.
The check it misses most: When runs keep falling short, you see the exact line they miss.
Sent back with the gap: Work that falls short goes back to the agent with what's missing, up to three tries.

Change it from the same chat.

Tell your agent what should change. A new instruction or checklist is saved as a new version, and the next run uses it.

Every version kept: See who changed what and when, and bring any version back.
No lost edits: A change someone made in Kavela meanwhile is never overwritten.
Ten runs before a verdict: No version is called better until it has 10 checked runs.

ChecklistVersion 2

reports/hn-ai-tldr.md exists and was written today (must pass)
It covers exactly 3 posts
Every post has its link
No post is older than 24 hours

kavela - update_workflow (MCP)(workflow: "Morning AI news TLDR", version: 1, checks: [4 lines])

Saved Morning AI news TLDR as version 2. Changed: the checklist. The next run uses it.

kavela - list_workflow_versions (MCP)(workflow: "Morning AI news TLDR")

Morning AI news TLDR:

- v2 (live), saved 2026-10-23 16:28 Asia/Singapore by the person.

- v1 (earlier), saved 2026-10-23 16:27 Asia/Singapore by the person.

Common questions.

Which agents can connect?

Any MCP client that connects over HTTP, including Claude Code, Codex, Cursor, Claude, VS Code, Windsurf, Cline, Zed and OpenCode.

How does it sign in?

Your agent opens a Kavela sign-in in your browser. You can also send a Kavela API key as a Bearer header, and a read-only key can list and read but never change anything.

Does anything run on my computer?

No. Kavela runs every workflow in the cloud. Your agent saves it, changes it and reads how it went.

What else can my agent do through it?

Hand Kav one-off work with chat_with_kav, read and write the workspace's files, and answer decisions waiting for you. A final sign-off, a password or taking over a live browser stays in Kavela.

Can another service start a workflow?

Yes. A workflow that starts on events gets its own web address. Its signing secret is shown once, to you, in Kavela, and never through the MCP.

Where do I learn to write a good workflow?

docs.kavela.ai covers what to put in the instruction, the checklist and the schedule, in Write a workflow.

How much does it cost?

Every plan starts with a 14-day free trial, no card. Bring your Claude or Codex subscription at no extra cost, or use the included Kav AI usage.

Connect your agent.

Connect MCP

14-day free trial, no card. Bring Claude or Codex, or try included Kav AI usage.