Playwright MCP browser context is lost after 8-12 minutes of idle time.
Hi Everyone, I'm building an AI agent that uses Playwright MCP to control a browser, and I'm running into an issue where the browser context is lost after the agent remains idle for a while. Architecture My current setup looks like this: AI Agent (Client) hosted on Azure Container Apps Playwright MCP Server hosted on Azure Container Apps Chrome Extension running in the user's browser The extension maintains a webhook connection to the Playwright MCP relay/server. The MCP server generates Chrome DevTools Protocol (CDP) commands, which are executed by the Chrome Extension in the user's browser. This allows the browser to continue running on the user's machine while the agent and MCP server are hosted remotely. Issue Everything works as expected while the agent is actively interacting with the browser. However, if the agent stays idle for around 8–12 minutes, the next request fails with an agent response that the browser context is lost (or the context/browser is no longer available). If I ask the agent to retry, it invokes the browser_close tool and creates a new browser session. While this allows execution to continue, it completely loses the previous browser state, including: Current page Navigation history Login/session state tied to that browser context Any work already completed in the existing session Ideally, I want the existing browser session and context to remain usable even after periods of inactivity. Questions Is there any idle timeout in Playwright MCP or its browser context management that could explain this behavior? Could this be related to the CDP connection between the Chrome Extension and the MCP server being dropped after inactivity? Are there recommended keep-alive or heartbeat mechanisms for long-lived browser sessions? Has anyone successfully maintained browser contexts for extended idle periods (30+ minutes or longer)? Is there a recommended pattern for reconnecting to an existing browser context instead of creating a new one when the connection is interrupted? Environment Playwright MCP Server hosted on Azure Container Apps AI Agent hosted on Azure Container Apps Chrome Extension executing CDP commands in the user's local Chrome browser Communication between the agent, MCP server, and extension over the network I'd appreciate any guidance on where to investigate this issue or best practices for maintaining persistent browser contexts in this kind of architecture. If anyone has encountered a similar problem, I'd be interested in hearing how you resolved it. Thanks in advance!Looking for Guidance: Improving Playwright MCP Runbook Replay Reliability
Hi everyone, We are building an AI-powered browser automation agent using Playwright MCP, and we have implemented a runbook/replay mechanism to reduce repeated reasoning and improve navigation performance. Current Architecture Our setup currently looks like this: AI Agent (Client) – Hosted on Azure Container Apps Playwright MCP Server – Hosted on Azure Container Apps Chrome Extension – Running in the user's browser The Chrome Extension maintains a webhook connection with the Playwright MCP relay/server. The MCP server generates Chrome DevTools Protocol (CDP) commands, which are executed by the Chrome Extension in the user's browser. Runbook Functionality When the agent needs to navigate to a particular page, it normally performs multiple browser actions such as: browser_navigate browser_snapshot browser_click browser_type Other Playwright MCP tools During the first successful execution, we record the tool calls and their parameters used by the agent to reach the desired page. We store these actions as a runbook. For subsequent requests requiring navigation to the same destination, instead of asking the AI agent to reason through the navigation steps again, we directly replay the stored runbook. First execution User Request ↓ Agent Reasoning ↓ MCP Tools ↓ Browser ↓ Desired Page Example recorded sequence: browser_navigate → browser_snapshot → browser_click → browser_snapshot → browser_type ... Subsequent execution User Request ↓ Runbook Replay ↓ MCP Tools ↓ Browser ↓ Desired Page This significantly reduces latency, token usage, and unnecessary LLM reasoning. Example Runbook Step { "tool": "browser_click", "args": { "target": "e140", "element": "Earn credentials link" }, "a11y": { "role": "link", "name": "Earn credentials", "index": 0, "text": "Earn credentials", "textIndex": 0, "path": "generic/list/listitem/article/generic/generic", "url": "https://learn.microsoft.com/en-us/" }, "pageUrl": "https://learn.microsoft.com/en-us/", "timestamp": "2026-07-30T10:24:04.374Z" } For each recorded action, we persist information such as: MCP tool name Tool parameters Snapshot reference Accessibility role Accessibility name Element text Element index Text index Accessibility-tree path Page URL Other relevant metadata The Problem The main challenge is runbook replay reliability. A runbook contains references obtained from a previous browser_snapshot. During replay, the page may have: Slightly different DOM structure Timing differences Dynamic content Accessibility-tree changes As a result, a previously recorded snapshot reference may no longer match the current page. For example, a recorded action may effectively be: browser_click(ref="some_snapshot_ref") During replay, that reference may no longer exist or may resolve to a different element because the current snapshot has changed. This causes intermittent failures in operations such as: browser_click browser_type Other element interactions Current Status We have improved the replay mechanism and currently achieve approximately 90% replay success. The remaining ~10% failures are difficult because they are often nondeterministic. The same runbook can succeed or fail depending on the current browser or page state. Questions We would appreciate advice from anyone with experience using Playwright MCP or building production browser automation systems. What is the recommended way to make Playwright MCP tool calls more resilient to changes in snapshot or accessibility-tree references? Is relying on snapshot ref values fundamentally unreliable for replay? Should we instead store a different representation of the target element? If so, what additional information would you recommend storing for each interaction? What is the best strategy for re-resolving an element when the original snapshot reference is no longer valid instead of failing the entire runbook? Are there Playwright MCP capabilities or recommended patterns that specifically address dynamic pages and changing accessibility trees? For a production-grade browser automation agent, what architecture or strategy would you recommend to achieve >99% replay reliability? Our goal is not to eliminate dynamic behavior from websites, but to make the runbook intelligent enough to recover when previously recorded element references become invalid. Any architectural recommendations, implementation patterns, or examples from production systems would be greatly appreciated. Thank you!From Visibility to Impact: Building Your Personal Brand with Smart, Cost-Effective Tools
Beyond mastering technology, MVPs are passionate about helping others grow, supporting both technical and soft-skills development, and creating opportunities for learning and collaboration. They actively share their discoveries, experiences, and lessons learned while engaging in valuable exchanges with fellow community leaders and influencers. In this spirit of continuous learning and knowledge sharing, building a personal brand becomes a powerful way to amplify your voice, expand your reach, and increase the impact you can have on the communities you serve. Recently, the global MVP community came together for a discussion on creating a personal brand, best practices, tips and effective tools that can be leveraged to raise your social media reach. The session titled “From Visibility to Impact: Building Your Personal Brand with Smart, Cost-Effective Tools,” featured insights from Femke Cornelissen, Microsoft 365 and Copilot MVP from the Netherlands, and Fabio Bonolo, Microsoft 365 Copilot MVP from Switzerland. Together, they explored one of the most valuable skills for today's community leaders and technology advocates: building a personal brand that creates meaningful impact while staying authentic and accessible. Through personal experiences, practical examples, and demonstrations of cost-effective tools, they showed MVPs how visibility can become a catalyst for community growth, professional opportunities, and wider knowledge sharing. Personal Branding Starts with Authenticity Opening the session, Femke Cornelissen shared her personal journey to becoming an MVP and the lessons learned along the way. One of her strongest messages was clear: authenticity matters more than following a formula. Rather than trying to replicate someone else's success, Femke encouraged attendees to identify their own passions and areas of expertise. Whether sharing insights on Microsoft technologies, community initiatives, innovation, or personal experiences, the most effective personal brands are built around genuine interests and perspectives. She highlighted how consistent visibility on platforms such as LinkedIn can create unexpected opportunities—from speaking engagements and community leadership roles to customer conversations and valuable professional connections. Yet success does not happen overnight. Building recognition requires showing up consistently, sharing valuable insights, and developing trust over time. LinkedIn as a Community Platform While many professionals still view LinkedIn primarily as a digital résumé, Femke explained how the platform has evolved into a powerful space for thought leadership, networking, and community engagement. She encouraged MVPs to think beyond job titles and focus on creating content that demonstrates their expertise and unique perspective. Personal stories, lessons learned, event experiences, and practical guidance often resonate more strongly than purely promotional content. According to Femke, people connect with people—not corporate messaging. Several practical recommendations emerged from her presentation: Develop a consistent presence rather than posting sporadically. Share insights in your own voice. Engage actively with other community members through comments and conversations. Support and elevate the work of others. Focus on meaningful engagement rather than follower counts. She also emphasized the importance of balancing personal, community, and professional content to maintain authenticity while still supporting employers, projects, and initiatives that align with personal goals. AI as a Productivity Multiplier The session also explored how AI can help content creators work more efficiently without losing their human voice. Femke demonstrated how she uses AI-powered workflows to monitor industry trends, draft content ideas, analyse performance per post, and manage large volumes of community interactions. Rather than replacing authentic communication, she views AI as a tool that enables her to spend more time building relationships and less time managing repetitive tasks. One particularly important takeaway was her approach to maintaining authenticity. While AI helps create first drafts and analyse data, final content should always reflect an individual's own perspective, experiences, and tone of voice. Community members follow people for their unique insights, not automatically generated content. Designing a Recognisable Personal Brand Building on Femke's focus on authenticity, Fabio Bonolo explored the visual and strategic side of personal branding. His presentation introduced a structured framework built around four key pillars: Identity, Content, Reach and Consistency. Fabio explained how visual identity contributes to recognition and memorability. Colours, logos, typography, imagery, and design choices all communicate aspects of a person's personality and expertise. By maintaining consistent visual elements across websites, presentations, blogs, videos, and social media channels, community leaders can create a stronger and more recognisable presence. He demonstrated how even simple visual standards can make content instantly recognisable while reinforcing credibility and professionalism. Powerful Tools Without Enterprise Budgets One of the most popular parts of the session was Fabio's walkthrough of affordable tools that MVPs can use to professionalise their online presence. Among the tools showcased were MojoMox for logo creation, colour palettes, and brand design systems, PowerPoint as a surprisingly powerful design tool for creating thumbnails, graphics, and consistent visual assets, AuthoredUp for enhancing LinkedIn content creation and analytics, Wix for website management, blogging, and email marketing, Clipchamp for video creation and editing as well as Riverside and StreamYard for podcasting, live streaming, and content recording. A recurring theme throughout Fabio's presentation was that impactful personal branding does not require expensive software or large marketing budgets. By combining a clear strategy with carefully selected tools, MVPs can create professional-quality content for a relatively small investment. Community Remains the Force Multiplier Despite the extensive discussion about tools, automation, AI, and branding strategies, both presenters arrived at the same conclusion: community is the most important element. Femke described personal branding, AI, and community as three interconnected forces that create impact when combined effectively. Visibility alone is not the goal; meaningful relationships, knowledge sharing, and helping others succeed are what ultimately build a lasting reputation. Fabio reinforced this message by encouraging attendees to stay true to their identity, create content that delivers value, and remain consistent across the channels where they engage their audiences. Key Takeaways for MVPs The session concluded with several practical lessons for anyone looking to expand their impact: Be authentic and share your unique perspective. Focus on helping others rather than chasing metrics. Develop a consistent personal brand across your channels. Use AI to enhance productivity, not replace authenticity. Invest in relationships and community engagement. Start with simple, affordable tools and scale over time. Stay consistent, visibility is built through sustained effort. As the MVP community continues to grow and evolve, sessions like this demonstrate that the most powerful brands are not built through marketing alone—they are built through expertise, generosity, authenticity, and community. Thank you to Femke Cornelissen and Fabio Bonolo for sharing their experiences and practical guidance, and to all MVPs who participated in the discussion and contributed their own insights throughout the session. Want to Learn More About the MVP Program? To find an MVP and learn more about the MVP Program visit the MVP Communities website and follow our updates on LinkedIn or #mvpbuzz. Join us for a future live session through the Microsoft Reactor where we walk through what the MVP program is about, what we look for, and how nominations work. These sessions are designed to help you connect the dots between the work you’re already doing and the impact the MVP Program recognizes - with time for questions, examples, and real conversations.
Is it possible to do this?
Hey all! I currently use Smartsheet to track my trainees' progress and I love the way I can use Conditional Formatting in that program. Unfortunately, my company is going away from using Smartsheet, so I need to move all the information I have to an Excel document. I'm struggling with getting the Conditional Formatting to work the same way I have it in Smartsheet and I'm curious if it's even possible. Here is an image of the Template I will be using for each employee. My plan is to have a tab for each employee, just duplicating the template and renaming it to their name and adding their information. Here are the two things I'm struggling with getting set-up with Conditional Formatting: 1. In the Lvl column, I have a drop down that is linked to the Data tab where the icons on the left are images. I want to be able to select the "Assigned" option and have the A icon show up (this is currently set up) and then have the fill of the box turn to red. I'm struggling to get that to work properly. Currently, when I select the "Assigned" option in the drop-down menu, it just shows the "A" icon. 2. I also want to be able to have the "Date Completed" column change to a specific color if one of the checkbox columns (G, H, I, J, K) are checked (i.e. if G is checked, it turns a light brown, if H is checked, it's gray, if I is checked, then it's red, etc.). This is what it looks like in Smartsheet and how I'm hoping to get it to look in Excel: Thank you!
How (if possible) do I remove thi
Hi. Windows Update wants to install PowerShell v7.3.1 (x64) on my desktop computer, even though I already have newer version, PowerShell 7.6.3.0-x64, installed! I have tried a lot of those 'tips' found with a Google search, but 10+ restarts later, I still have the view seen on WU.jpg! Is there any way I can remove that annoying message? Have a nice day :-) WU.jpg WU2.jpg
Windows 11 Home - fresh install - bitlocker problem
Hello i just did a fresh install of windows 11 home on my desktop. And after rebooting my second ssd ( partition d ) is asking for a bitlocker key. I can't connect to https://aka.ms/recoverykey the site dosen't work for me. Is there a way to recover my data or it is lost forever?
