As part of our continuing investigation into the evolution of developer tools, we had a conversation with Ostap Elyashevskyy, our Test Automation Competence Manager, to learn more about how new developments are changing the software development environment.
Two of GitHub Copilot's capabilities, Agent Mode and Model Context Protocol (MCP), stand out in particular as the platform develops beyond its original potential. Ostap gives a summary of these features and explains how they improve workflows for testing and development.
Can Agent mode and Model Context Protocol elevate the GitHub Copilot Experience?
Yes, Agent mode and MCP servers can significantly enhance the GitHub Copilot experience. Before the support of Agent mode, Copilot was a more autocomplete tool which required step-by-step commands of what to do. Now Agent mode + MCP makes it a more autonomous and goal-seeking tool rather than a declarative tool. Instead of typing “write a test for this function,” you could say, “Ensure this module is fully tested and bug-free,” and it figures out what to do. Moreover, MCP servers give your agent context and tools, enabling interactions with applications under test at a completely new level. It’s like the missing puzzle piece that’s now available in GitHub Copilot or Cursor.
To summarise, let’s review how Agent mode and MCP improve GitHub Copilot:
| Feature | Agent Mode | MCP Servers | Description |
|---|---|---|---|
| Understands goals | ✅ | ❌ | Copilot can follow high-level tasks like “migrate to Playwright.” |
| Plans multi-step tasks | ✅ | ❌ | It breaks big tasks into smaller steps and completes them. |
| Access to full project context | ❌ | ✅ | It can see all code files, configs, docs, and test results. |
| Uses external tools (linters, tests) | ❌ | ✅ | Copilot can run tests, check formatting, or call dev tools. |
| Remembers project actions | ✅ | ✅ | Keeps track of what it's done to make smarter future changes. |
| Flexible, plugin-based system | ❌ | ✅ | You can add tools or agents (e.g., a security checker) easily. |
Together, they form the backbone for the next-gen Copilot, capable of resolving problems in agentic mode without much user intervention.
How can MCP impact the testing process?
Model Context Protocol is a game changer as it adds context to your prompts: for example, let's say that we have a Web application and we must write tests based on test cases. We use Playwright MCP Server, which knows how to deal with web applications, knows how to parse the DOM model and can generate valid tests or scripts which interact with web elements. It’s like a bridge to the application under test.
There are dozens of MCP servers for different purposes. For example, for testing, the most helpful are Web, API, DB, Accessibility, Security, Mobile testing, MCP servers and others. It is worth mentioning that usage is broad: from generation and execution of code in agentic mode to asking specific questions about DB schema, or fields in API requests, locators for some complex elements, and so on. It means that Manual and Automation test engineers can utilise the MCP servers' concept in their work. Together, the agentic mode and the MCP make Copilot more powerful than it was before. It is often referenced as the “USB-C” of AI agents, meaning that it allows connecting different AI agents through a common interface.
What are the benefits of MCP in testing?
- Dynamic test generation. It allows you to generate/execute code based on the context of your application.
- Maintenance. If a model context changes, for example, data type of some field in the REST request or UI locator of the page, it is easier to update tests as the MCP server has access to the context.
- Quick review and feedback. For example, we want to find broken links, do an evaluation of accessibility issues on the page, do API tests for the given endpoint, security check of the login form, check coverage according to requirements, and many other cases where it is helpful.
- Risk-based testing, better test coverage. Having access to context provides more information to LLM about risky areas and edge cases. This helps to generate more relevant tests or suggestions about what to test.
- Other benefits which are not so obvious from the first perspective, for example, faster onboarding–you can ask questions about the application, its architecture and interfaces directly in the GitHub Copilot with MCP servers. If you are learning DB structure to create SQL requests, you can ask about tables, their relation, fields/format to better understand how they are connected, and finally generate/explain a SQL query to fetch needed data.
Are there any security concerns with MCP?
Security is the most important concern in the MCP architecture. Because it allows agents to freely share memory and invoke tools without strict checks, a compromised tool can corrupt the system, leak sensitive data, or trigger widespread failures. Without stronger security controls like access restrictions, validation, and versioning, MCP can become a serious vulnerability.
- To secure MCP, you should take the following actions:
-
- Control access. Only trusted agents and tools should be allowed to access or change memory and actions. For example, GitHub Copilot or Cursor with an AI Assistant downloaded from the official websites are trusted agents.
- Validate. Always check and verify any information agents share before accepting it or acting on it.
- Isolate sensitive data. Keep critical systems separate so a compromised agent can't reach everything at once.
- Monitor for abnormal behaviour. Review agents for unexpected actions, like accessing tools they normally shouldn’t use.
Related insights
