Agent mode tools in VS Code  ## Metadata - Author: [[Visual Studio Code]] - Full Title: Agent mode tools in VS Code - Category: #articles - Document Note: --- tags: [#VSCode, #Copilot, #AgentMode, #AI, #DeveloperTools, #MCP] - #VSCode - #Copilot - #AgentMode - #AI - #DeveloperTools - #MCP Author: [YouTube Presenter, GitHub Copilot Team] Source: [YouTube: Agent mode tools in VS Code](https://www.youtube.com/watch?v=VePxCcF99w4) Reference: [Transcript, YouTube Video](https://www.youtube.com/watch?v=VePxCcF99w4) date created: 2025-05-15 14:35 --- # Title: 🧑‍💻 Agent Mode Tools in VS Code: Supercharging Copilot with Extensions, MCP, and Workflow Automation ## Table of Contents: - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📖 References](#-references) - [🗣️ Paraphrase](#-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-diversity-equity-inclusion-and-belonging-assessment) - [✅ Logical Validity](#-logical-validity) - [🔍 DARVO Analysis](#-darvo-analysis-prompt) - [🔍 SEO Evaluation](#-seo-evaluation) - [🎯 Conclusion](#-conclusion) --- ## 📝 Summary: Agent mode in Visual Studio Code (VS Code) dramatically expands GitHub Copilot's power by enabling the AI to not only edit code but also invoke a wide array of tools, including built-in features, extension-driven helpers, and Model Context Protocol (MCP) servers. This workflow automation turns Copilot into a hands-on coding agent capable of searching your codebase, running commands, fetching web data, and integrating third-party AI-powered tools—drastically improving developer efficiency and customizability. --- ## 📖 Main Content: Agent mode transforms Copilot from a passive code suggester into an active assistant that can both act on your code and trigger specialized tools. Tools can originate from three main sources: 1. **Built-in Copilot Chat Tools**: Reference your codebase, fetch web data (#fetch), or provide contextual help. 2. **VS Code Extensions**: Extensions like "prompt boost" contribute new tools for prompt enhancement and better responses. 3. **MCP Servers**: The Model Context Protocol (MCP) standardizes how external AI models or tools provide context, improving integration and custom workflows. MCP servers are configured in the `MCP.json` file and can include user-secured variables (like API keys) for safe use. Agent mode enables users to: - Manage which tools are available via a tools icon UI in VS Code chat. - Invoke tools explicitly (e.g., with `#fetch URL`) or let Copilot auto-invoke as needed. - Customize trust and confirmation flows for tools, reducing friction for frequent trusted actions. - Seamlessly integrate results from tools like Perplexity, which can research or reason about development questions and edit your code accordingly. This mode turns VS Code into a hub for agentic, context-rich, AI-augmented development—speeding up routine workflows and empowering developers to build their own toolchains. ### 🗒️ Notes - *Agent Mode Overview* - Empowers Copilot to actively edit code and trigger tools. - Access and manage tools via the agent mode chat interface. - Tools include searching the codebase, running terminal commands, and reading files. - *Types of Tools in Agent Mode* - **Built-in Tools**: Native Copilot features (codebase search, #fetch for web data). - **Extension Tools**: e.g., "prompt boost" to refine prompts and embed technical best practices. - **MCP Tools**: Via Model Context Protocol servers, offer advanced, external, or organization-specific capabilities. - *Tool Management and Workflow* - Users can toggle available tools in agent mode. - Tools invoked by explicit hashtags or natural language (e.g., "fetch from webpage"). - Extensions can register new tools for agent mode use. - Copilot can auto-invoke extension/MCP tools as part of the chat workflow. - *Security and Configuration* - MCP servers configured in `MCP.json`, including secure input variables (API keys). - VS Code prompts for sensitive data securely on first use, storing it safely. - *Automation and Trust* - Tools can require confirmation; frequent tools can be "auto-approved" to avoid repeated prompts. - Results from tools like Perplexity feed directly into code, maintaining a smooth developer workflow. - *Extensibility and Customization* - Developers can build their own tools using the language model tool API. - MCP support enables virtually limitless integration with external services. - *Call to Action* - Users encouraged to explore agent mode, build custom workflows, and share feedback or topic requests in comments. --- ## 📖 References: - [Agent mode tools in VS Code (YouTube)](https://www.youtube.com/watch?v=VePxCcF99w4) - [VS Code Documentation](https://code.visualstudio.com/docs) - [GitHub Copilot](https://github.com/features/copilot) - [Model Context Protocol (MCP) on GitHub](https://github.com/mcp-wp) - [Perplexity AI](https://www.perplexity.ai/) --- ## 🗣️ Paraphrase: 1. **Original:** "Agent mode in VS Code is like magic. In agent mode, Copilot can make edits to your code and invoke tools to help accomplish the request that was made. In today's video, I'm going to talk a bit more about what exactly tools are and the different ways that they can be provided." **Paraphrased:** Agent mode transforms Copilot from a simple assistant into an active collaborator—it can directly modify your code and activate specialized tools to fulfill your requests. This video explains how these tools work and the various methods of making them available in VS Code. 2. **Original:** "Extensions can contribute tools to agent mode. So if we head over to agent mode and click on the tools icon, you will see the prompt boost tools from the prompt boost extension. So I can reference this tool by typing #prompt boost. But the real power comes from agent mode invoking it automatically based on the user's prompt." **Paraphrased:** VS Code extensions can add their own tools to agent mode. For example, after installing the prompt boost extension, its tools appear in the agent mode menu and can be triggered using #prompt boost—or, better yet, agent mode can automatically run them when your conversation suggests it's needed. --- ## 📰 Newsletter: **🚀 Supercharge Your Coding: VS Code’s Agent Mode Unleashes Copilot’s Power** Hey devs! This week we’re diving into VS Code’s new Agent Mode—a game-changer for anyone using GitHub Copilot. With Agent Mode enabled, Copilot isn’t just suggesting code—it can edit your files and trigger an expanding universe of tools to do everything from web research (#fetch) to managing your workflow with custom extensions like Prompt Boost. Want more AI? The new MCP server support means you can hook in powerful tools (like Perplexity AI) right inside VS Code—no more tab switching, just smarter coding. Plus, setup is secure with variable management for API keys and robust automation options. Check out the full walkthrough on YouTube, and let us know: which tools or integrations should we cover next? 🔗 [Watch the video](https://www.youtube.com/watch?v=VePxCcF99w4) | 🧑‍💻 [Docs](https://code.visualstudio.com/docs) | 🤖 [MCP Protocol](https://github.com/mcp-wp) --- ## ✅ Action Items: - Share the Agent Mode video with dev teams exploring AI-augmented coding or workflow automation. - Review the tools enabled in your own VS Code agent mode and test extension/MCP integrations. - If onboarding new contributors, demo the agent mode UI to show how Copilot can boost productivity with real-time tool invocation. - For those working in DevRel, document real-world use cases where agent mode solves common developer pain points (e.g., integrating security scans, accessibility checks, or automating documentation). - Align agent mode adoption with open-source advocacy goals—highlight how tool extensibility promotes community solutions and knowledge sharing. - As an aspiring OSPO director, promote secure practices (e.g., API key handling in MCP), ethical AI usage, and inclusivity in tooling documentation and workflows. --- ## 🧪 Hypotheses: - Integrating agent mode tools in VS Code will significantly reduce context switching and increase developer productivity. - Custom extension tools and MCP integration will drive higher adoption of Copilot among enterprise teams. - Secure handling of sensitive input variables (like API keys) via agent mode boosts organizational trust in AI-powered workflows. ### 📊 Methodology: - **Time-on-task studies**: Compare completion times for coding tasks with and without agent mode/tool use. - **Surveys/interviews**: Gather feedback from developers on usability, learning curve, and productivity impact. - **Telemetry analysis**: Track feature adoption rates and error rates for MCP tool invocations. - **A/B testing**: Experiment with different extension/MCP tool configurations to optimize workflow. ### ❓ Questions: - How does agent mode change the day-to-day workflow of VS Code users? - What are the most common pain points in configuring or trusting third-party tools/extensions in agent mode? - Which types of tools (code search, fetch, AI research, prompt boosting) are most valuable to developers, and why? - How can the community contribute new, inclusive tools for agent mode? - What security practices are critical when automating workflows with sensitive data in agent mode? ### 🔮 Future Research: - User experience studies comparing Copilot’s agent mode with alternative AI coding assistants. - Benchmarking MCP server integrations across different use cases (e.g., accessibility, security, research). - Impact analysis of agent mode on open-source contribution rates and project onboarding. - Development of best-practice guidelines for ethical and secure tool invocation in agentic workflows. ### 💡 Implications: - Mainstreaming agent mode and tool-based automation may redefine coding best practices, lowering the barrier for advanced workflows. - MCP support can serve as a blueprint for cross-platform, open integration of AI tools—fostering community innovation. - Security and ethical considerations around tool automation (API key management, confirmation flows) set important precedents for future AI/DevOps integrations. --- ## 🧠 Learning Styles Assessment: - **Readability:** Conversational video script; ~Grade 9-10 reading level. - **Visual Learners:** *Present:* Video with interface demos and iconography. *Improve:* Supplement with infographics, flowcharts, or step-by-step guides in docs. - **Aural Learners:** *Present:* Video narration. *Improve:* Offer a podcast version or transcript with audio highlights. - **Verbal Learners:** *Present:* Clear explanations and examples. *Improve:* Add written tutorials, analogies, or deeper technical breakdowns for textual learners. - **Physical Learners:** *Present:* Encourages hands-on exploration in VS Code. *Improve:* Provide downloadable sample projects or interactive walkthroughs. - **Logical Learners:** *Present:* Outlines workflows, settings, and configuration logic. *Improve:* Add troubleshooting guides, decision trees, or checklists. - **Social Learners:** *Present:* Prompts users to comment and suggest topics. *Improve:* Foster peer-to-peer forums, group hackathons, or office hours for shared learning. - **Solitary Learners:** *Present:* Can follow video individually. *Improve:* Add self-paced challenges or reflection prompts. - **Naturalistic Learners:** *Absent.* *Improve:* Draw analogies to natural systems or include sustainability discussions in future tool reviews. --- ## 📊 Accessibility Assessment: - **Screen Reader Compatibility:** Video transcript provided; ensure all visual demos are described in text. Docs and UI should use semantic HTML. - **Keyboard Navigation:** VS Code is generally accessible, but custom tools should be tested for keyboard-only operation. - **Contrast and Readability:** Depends on VS Code theme and video captions. Recommend high-contrast captions and clear visual demos. - **Text Resizing:** Video and transcript content is scalable on most platforms. - **Alt Text:** All screenshots and UI demos should include alt descriptions in documentation. --- ## 🌍 DEIB (Diversity, Equity, Inclusion, and Belonging) Assessment: - **Cultural Relevance:** Video uses US-centric tech jargon; *improve* by adding context for international developers and supporting i18n in tool docs. - **Inclusive Language:** Generally neutral and accessible; *improve* by clarifying terms for non-native speakers. - **Neurodiversity Support:** Video pacing is reasonable; *improve* by breaking tutorials into shorter, stepwise modules. - **Representation:** Lacks visible diversity in presenters; *improve* by featuring contributors from a variety of backgrounds in demos and success stories. - **Psychological Safety:** Tone is supportive and encourages community feedback. - **Trigger Warnings:** Not applicable. --- ## ✅ Logical validity: - **Sound Deductive Reasoning:** - The video logically connects Copilot’s agent mode capabilities with productivity benefits, backed by clear workflow examples. - **Strong Inductive Reasoning:** - Shows concrete use cases (e.g., using #fetch or prompt boost) to support claims about flexibility and efficiency. - **Consistent Analogies:** - Refers to agent mode as “like magic”—a simple, accessible analogy for non-technical audiences. - **Balanced/Nuanced View:** - Explains pros and security steps (e.g., input variables for API keys), balancing utility with safe practices. - **Logical Flow:** - Follows a clear introduction-to-practical-examples progression. - **No major fallacies detected.** Avoids overhyping, addresses limitations (security, configuration). - **Tone:** - Upbeat, educational, and friendly. Direct, active, and inviting to a wide developer audience. --- ## 🔍 DARVO Analysis Prompt - **Deny:** - No denial of limitations; instead, acknowledges setup and security steps. - **Attack:** - No personal attacks or negative framing. - **Reverse Victim and Offender:** - Not present; the content is solution-focused, not adversarial. - **Additional Notes:** - Emotional tone is enthusiastic and welcoming, encouraging participation and learning. - No DARVO tactics, manipulation, or adversarial rhetoric present. --- ## 🔍 SEO Evaluation 1. **Keywords:** - "Agent mode," "VS Code," "Copilot," "MCP," "tools," "AI coding assistant"—all present and relevant. 2. **Meta Description:** - Not shown in video, but recommended: “Explore how Agent Mode in VS Code empowers Copilot with powerful tool integrations, extensions, and secure MCP servers for next-level coding workflows.” 3. **Title Tag:** - Video and page titles include “Agent mode tools in VS Code,” covering primary keywords. 4. **Headers:** - Add section headers to docs/blog posts for improved SEO and skimmability. 5. **Content Length:** - Video/transcript length is appropriate; written summaries could expand for deeper engagement. 6. **Internal/External Links:** - Reference docs, MCP protocol, and Copilot info are included. 7. **URL Structure:** - YouTube link is short and keyword-rich; use similar structures in blog docs. 8. **Image Optimization:** - For blog/website use: ensure screenshots have descriptive alt text and filenames. 9. **Mobile Optimization:** - YouTube and VS Code docs are mobile-friendly. 10. **User Engagement:** - Video encourages comments; add share buttons and related articles to blog. 11. **Schema Markup:** - Use VideoObject schema for YouTube embeds; FAQ or HowTo for tutorial docs. 12. **Social Sharing:** - Open Graph and Twitter Card tags recommended for articles and docs. --- ## 🎯 Conclusion: VS Code’s Agent Mode unlocks a new dimension for Copilot, letting developers seamlessly blend coding, automation, and AI-powered research right in their editor. By combining built-in, extension-driven, and MCP-integrated tools—with a focus on security, flexibility, and accessibility—Agent Mode sets the stage for a smarter, more collaborative future in software development. Developers, advocates, and educators alike should keep an eye on these evolving workflows as they push open source and AI collaboration into exciting new territory. --- - Document Tags: [[mcp]] [[vs code]] - Summary: Agent mode can invoke tools to help accomplish a user's request - but what exactly does that mean? In this video, we'll go over what agent mode tools are and how they can be provided in VS Code. 00:00 Intro 00:16 What are tools? 00:42 Built in tools 01:24 Tools from extensions - Prompt Boost 02:25 MCP tools - Perplexity MCP server 🔗 Links: Agent mode blogpost: https://code.visualstudio.com/blogs/2025/04/07/agentMode Prompt Boost extension: https://marketplace.visualstudio.com/items?itemName=chrisdias.promptboost Language Model Tool API docs: https://code.visualstudio.com/api/extension-guides/tools MCP demo repo: https://github.com/digitarald/mcp-frontend-starter - URL: https://www.youtube.com/watch?v=VePxCcF99w4 Agent mode in VS Code is like magic.In agent mode, Copilot can make edits to your codeand invoke tools to help accomplish the requests that weremade.In today's video, I'm going to talk a bit moreabout what exactly tools are and the different ways thatthey can be provided.Tools in agent mode help Copilot get the job doneby providing its own additional capabilities, like searching the codebase, running a terminal command, reading files, and so muchmore. You can see your list of tools by switching toagent mode in Chat, then clicking the tools icon.From here, you can deselect or select which tools youwould like to make available to Copilot.Tools can come from built-in tools which are listedunder GitHub Copilot Chat and include things like codebase,which references information in your code base, or fetch, whichwill fetch web data and provide it as context. You can always reference these tools with a hashtag.So, for example, I can say "add Tailwind CSS #fetch"and then the URL.Or I can use natural language and say "fetch fromweb page," and then the URL, and it will automaticallycall the tool. This is where the power of agent mode really lies.Not only can it edit your files, but it caninvoke these tools to help you get the best possibleresponse.Now, there are two other places that tools can comeFrom extensions and MCP servers. With the Prompt Boost extension,let's first look at how extensions can provide tools. The Prompt Boost extension enhances your prompts to get better,more accurate responses by adding relevant technical context, best practices,and specific requirements.So with this extension installed, let's look at this markdownfile that I have with a simple prompt to createa to-do app.And this is the typical extension experience that you're usedto in VS Code, where it adds a feature to theeditor.So I can click boost prompt, and it will returna much more detailed prompt. But it would be really great if we could boostour prompt while we were talking to Copilot.And you can because extensions can contribute tools to agentmode.So if we head over to agent mode and clickon the tools icon, you will see the Prompt Boosttools from the Prompt Boost extension.So I can reference this tool by typing hashtag promptboost, but the real power comes from agent mode invokingit automatically based on the user's prompt.So if I type boost this prompt, then agent modewill automatically invoke it.Now we get our enhanced prompt returned that we canthen use to continue our conversation to get better resultswith a better prompt.This is a really great example of how the extensibilitythat you know and love in VS Code works withAI as well. And you can build your own. Check out our docs on the Language Model Tool APIto learn more. OK, so next I want to explore how agent modecan invoke MCP tools.MCP stands for Model Context Protocol and is an openprotocol that standardizes how applications provide context to models.And remember, better context equals better answers.In VS Code, MCP support enhances agent mode by allowing you to connectany MCP server and use the tools it provides. In agent mode, MCP server configuration happens in themcp.json file located in the .vscode folder.I already have a perplexity server configured here, but toadd a new server I could click the Add serverbutton directly in the JSON file where you can thenselect a type and it will add a template foryour new server. Now if we look at this perplexity server configuration, youwill see that we have general properties for setting thecommand and arguments.But then you will also see a property for aninput variable for our API key.Input variables let the user input sensitive information so youcan avoid hard coding it. And then we can reference that input in the perplexityAPI key property.When I go to start the server for the firsttime, inputs are prompted then stored securely by VS Code.So this is a really great feature that allows youto keep your secret safe. And we can run the MCP List Servers command tosee the list of servers I have and start orstop them.Now if we look at our tools available inagent mode, we see the Perplexity Ask, Research, and Reasontools from our Perplexity MCP server.I have this web page running and it's all aboutMCP.So let's add a page about the top MCP servers.If I prompt "research with Perplexity what the best MCPservers for front-end development are and create a splashylisting page," it will automatically invoke the Perplexity research tool. Here's one little hack you might not know about.In agent mode, when certain tools are invoked, Copilot requestsconfirmation to run the tool.If you have a tool that runs frequently and thatyou trust, you can use the continue button drop-downoptions to automatically confirm the specific tool for the currentsession, workspace, or all future invocations.Then you won't get prompted the next time it runs. When I continue with this tool, we have the powerof Perplexity at our fingertips.This will take a few moments, so I did makesome quick editing cuts here so we can skip ahead.Once the tool call is finished, you can see theoutput from it, so we can see the full responsethat Perplexity returned.Then Copilot will take that result and edit our codeto implement our new page, all without us having toleave our agentic coding flow. Agent mode is powerful, and there are so many toolsyou can leverage to get the best possible results foryour workflow.Have you started coding with agent mode?Let us know in the comments, and if there areany topics you’d like us to explore in future videos.Happy coding!, Agent mode tools in VS Code, Visual Studio Code, 01jvam8pnyrqmr0qb1bmcbyqkc.