Skyvern

@Skyvern list all agents in the 'customer-onboarding' folder and show me which ones have run in the last 7 days

Found 4 agents in the 'customer-onboarding' folder. Two have recent activity: 'Verify Email Agent' (last run 2 days ago, completed successfully) and 'Document Upload Agent' (last run 5 days ago, failed with timeout). The other two agents—'Welcome Survey Agent' and 'Payment Setup Agent'—have not run in the past week. All agents are on their latest versions.

This example combines folder lookup with run history filtering to surface agent usage patterns. Requires read access to both agents and runs. The MCP returns metadata only; you won't see the actual browser automation steps or screenshots unless you fetch a specific run's observation links.

@Skyvern cancel all runs that have been active for more than 30 minutes and show me which workflows were affected

Canceled 3 active runs that exceeded the 30-minute threshold. Affected workflows: 'Invoice Processing' (2 runs canceled, both stuck at login step) and 'Data Entry Workflow' (1 run canceled, mid-form submission). All cancellations succeeded. No runs under 30 minutes were touched. You may want to investigate why the Invoice Processing workflow is stalling at authentication.

This demonstrates bulk cancellation with filtering logic applied by the AI. Skyvern's Cancel Runs tool accepts multiple IDs in one request, but the MCP doesn't auto-detect 'stuck' runs—you define the criteria in your prompt. Canceling a run stops browser automation immediately and may leave external systems in incomplete states.

@Skyvern create a new folder called 'Q1-2025-Experiments' and move all agents tagged 'prototype' into it, then summarize what got moved

Created folder 'Q1-2025-Experiments' (ID: fld_x7k2m). Moved 5 agents tagged 'prototype': 'LinkedIn Scraper v2', 'Form Filler Test', 'Multi-Step Checkout', 'PDF Extractor Beta', and 'API Fallback Handler'. All agents remain at their current versions and retain their run history. The folder now contains 5 agents; no other folders were modified.

This example shows folder creation followed by agent organization. The MCP's Create Folder and List Agents tools don't directly support 'move' operations—the AI infers which agents match your criteria and you'd update them separately (Skyvern's API handles agent-folder association outside these read/write primitives). Useful for keeping experimental automations separate from production workflows.

When Skyvern beats manual scraping for weekly price checks

A 6-person e-commerce team needs to track competitor pricing across 20 sites every Monday morning. Skyvern wins here because the agents handle site structure changes without brittle selectors, and the folder system keeps pricing agents separate from checkout-testing agents. Use List Agents to audit what's deployed, then List Runs with status filters to catch failures before the pricing deck goes to leadership. The trade-off: if your competitors block automation or require CAPTCHA solving, Skyvern's agent approach still hits walls. This works when you're scraping public catalog pages at low frequency and can tolerate a 2-hour refresh window. If you need sub-minute pricing updates or are scraping login-gated content, evaluate whether the API_KEY scope covers your auth needs first.

How Skyvern fits end-to-end testing for small product teams

A 4-person SaaS team ships checkout changes weekly and wants to catch regressions before customers do. Skyvern is borderline here. The Get Run tool surfaces observation links and error logs, which beats screenshot-only tools when a test fails. Create Folder keeps staging agents separate from production monitors. But if your checkout has more than 8 steps or requires dynamic data (promo codes, inventory checks), you'll spend more time debugging agent runs than writing Playwright scripts. This makes sense when your flow is stable, changes are cosmetic, and you want to avoid maintaining selectors across redesigns. If your team already writes code-based tests or your checkout logic changes every sprint, stick with Playwright and save Skyvern for the competitor-monitoring use case.

When Skyvern handles repetitive form-filling for support teams

A 3-person support team processes 40 vendor RMA forms per week, each requiring 12 fields copied from Zendesk tickets. Skyvern wins when the forms are on third-party sites you don't control and the fields rarely change. Use List Runs to spot submission failures, then Cancel Runs to stop duplicate attempts when a vendor site goes down mid-batch. The folder system organizes agents by vendor, so onboarding a new RMA process doesn't pollute your existing setup. The limit: if the forms require file uploads from varied sources or multi-step approvals, the 10-tool API surface might not expose enough control. This works when the task is copy-paste-submit and you're doing it more than 10 times a week. Below that threshold, manual entry is faster than maintaining agents.