Efficient Workflows: Automation, Browser, and MCP
💡 Goal: 3 lessons of about 30 minutes each, building the three-piece toolkit of the OpenClaw workflow—scheduled tasks let Claw work on schedule, browser tools go online for you, and MCP gives it an external toolbox. After each lesson, you'll have a finished result you can verify on the spot.
The first two stages solved "installed properly" and "chatting smoothly": the model is connected, the commands are learned, and Claw remembers you now. But the moment you step away, it stalls—that's not efficiency.
What the OpenClaw workflow tackles is the next step—getting Claw to get things done while you're not watching: automatically summarizing overnight news at 7 in the morning, speaking up the moment a webpage you're watching updates, or plugging in an MCP server whenever a tool is missing. This stage uses 3 lessons to thoroughly cover each piece of the "automation, browser, MCP" toolkit.
🎯 Stage Goals
- Build your first scheduled task:
openclaw automations listcan show it, it runs automatically at the appointed time, and results are delivered to a chat channel or webhook endpoint - Let Claw go online for you:
web_searchto search for information,web_fetchto fetch webpage content; when buttons need clicking, forms filled, or screenshots taken, hand it to the isolatedopenclawmanaged browser - Connect your first MCP server: add it in Control UI's Settings → MCP → Add server, verify connectivity with
openclaw mcp doctor <name> --probe
⏳ Estimated Time
- 3 lessons totaling about 30 minutes; about 21–39 minutes with buffer time reserved
- Each lesson stands on its own and can be read individually; 3.1 scheduled tasks are used most frequently, so we recommend starting there
- In a hurry for results: after finishing 3.1 and 3.2, you can head straight to the daily morning report and webpage monitoring in Scenario Practice—those two pages are exactly the combined application of these two lessons
🚀 Before You Start
- You've completed the Daily Use stage: Gateway is running, at least one model is connected (
/model statuscan report the current model) - You know how to open Control UI (
openclaw dashboard, local defaulthttp://127.0.0.1:18789/)—3.3 adding an MCP server takes a few clicks inside it - Want scheduled tasks to send results into IMs like Feishu or WeCom? Channel integration is in the Scenario Practice stage—not connecting doesn't affect this stage; tasks run all the same, and run history can be checked anytime with
openclaw automations runs <job-id> - A cup of water ☕—the goal of all three lessons is "build something that runs," and every command is short
📚 Lessons in This Stage
First, let's answer a selection question: "I want Claw to do X—which mechanism should I use?" The table below is compiled from the official Automation selection guide and tool overview; find your match, then enter the corresponding lesson:
| You want Claw to… | Which mechanism | Corresponding lesson |
|---|---|---|
| Send a report on time every morning at 9 | Automations (--cron recurring) | 3.1 Scheduled Tasks |
| Remind me of something in 20 minutes | Automations (--at one-time) | 3.1 Scheduled Tasks |
| Trigger the moment an external service sends an event (e.g., a code push) | Webhooks (POST /hooks/agent) | 3.1 Scheduled Tasks |
| Safely trigger processing when a new email arrives | IMAP plugin (sender gating) | Scenario Practice: Email Assistant |
| Search for information, fetch the content of a webpage | web_search / web_fetch | 3.2 Browser Control |
| Actually operate webpages: click buttons, fill forms, take screenshots | browser tools (managed browser profile) | 3.2 Browser Control |
| Borrow tools from another program | MCP servers (mcp.servers) | 3.3 MCP Integration |
| Run scripts when a session resets, enforce a rule before every reply | Hooks / Standing Orders | Advanced mechanisms, not covered in this stage |
Lesson schedule:
| Lesson | Content | Duration |
|---|---|---|
| 3.1 Scheduled Automations: cron, reminders, and Webhook triggers | Five scheduling methods (--at, --every, --cron, --on-exit, --stream-command), delivering results to chat channels, inbound Webhooks, the day/weekday OR pitfall in cron expressions | 12 minutes |
| 3.2 Browser control: let Claw browse the web for you | web_search / x_search / web_fetch lightweight trio, an isolated openclaw managed browser, reusing your logged-in Chrome session via the user profile | 10 minutes |
| 3.3 MCP integration: giving Claw an external toolbox | What MCP is, three integration surfaces (Settings / CLI / config file), reverse exposure via openclaw mcp serve, integration does not bypass tool policies | 8 minutes |
A quick look first: what the trio looks like assembled
- Daily morning brief: a scheduled task (
--cronset for the morning) +web_searchto gather news +--announce --channelto deliver to your IM → Scenario in practice: daily morning brief - Web page monitoring: a condition-based trigger that watches a target page on a schedule +
web_fetch/ browser to fetch content, notifying you only on changes → Scenario in practice: web page monitoring - GitHub on-call: inbound Webhooks wake the agent on PR and issue events, producing PR summaries and handling issues automatically → Scenario in practice: GitHub on-call
💡 Why do these three lessons come before the scenario practice?
The finished products in the scenario practice—the morning brief, monitoring, on-call—are all, when taken apart, parts from these three lessons: a scheduling or trigger mechanism, a pair of hands for browsing the web, and one or two add-on tools. Get the parts in hand first, and assembly becomes a matter of following along.
🎒 What you'll gain
- A Claw that "does the work on schedule": hands-on experience with all three wake-up methods—cron loops, one-off reminders, and Webhook triggers—with results automatically delivered wherever you specify
- A pair of hands that browse the web for you: the lightweight trio handles searching and fetching, the isolated managed browser handles clicking, form filling, and screenshots, without touching your own browser
- A plug-and-expand toolbox: MCP servers can be integrated from any of the three surfaces—Settings, CLI, or config file—so Claw's capabilities are no longer limited to built-in tools
- An assembly roadmap: the morning brief, web page monitoring, and GitHub on-call in Scenario practice are all direct combinations of parts from these three lessons
Once you finish these three lessons, the OpenClaw workflow upgrades from "you ask, it answers" to "it gets things done for you": what needs scheduling gets scheduled, what needs the web goes on the web, and missing tools get added as plug-ins. Next, go to Scenario practice to assemble what you want; if you'd rather catch up on Skills and security hardening first, the advanced handbook can be slotted in anytime.