GitHub Copilot Adds Vibe Coding Abilities

GitHub Copilot recently announced a significant update with the rollout of agent mode with Model Context Protocol (MCP) support to all Visual Studio Code users. This new mode, coupled with the integration of the MCP, aims to provide a more proactive and context-aware coding assistance experience.

The agent mode represents a shift from reactive code completion and multi-file editing towards a more action-oriented approach. It is designed to interpret user prompts and execute necessary subtasks to achieve a desired outcome. It seems like Microsoft is embracing the concept of ‘vibe coding’, which is better late than never.

This can include identifying and generating relevant files, suggesting and requesting the execution of terminal commands or tool calls, and analysing runtime errors with self-correction capabilities.

Support for the MCP was added to enhance agent mode’s capabilities. MCP allows agent mode to use external tools and information sources, providing a broader task context. GitHub also released a local, open-source MCP server, enabling the integration of GitHub functionalities into other LLM-powered tools that support the protocol.

With the agent mode, users can choose the model they want, including Claude 3.5 and Claude 3.7 Sonnet, Google Gemini 2.0 Flash, and OpenAI GPT-4o, which are included with all paid Copilot subscriptions. 

GitHub Copilot is introducing premium requests for newly available models used in chat, multi-file edits, and agent mode, supplementing the unlimited access to the base model for paid users.

Starting May, existing Copilot Pro, Business, and Enterprise subscribers will receive a monthly allocation of these premium requests, with unlimited access until then. A new Pro+ plan for individuals offers a higher allocation for $39 per month. All paid users will also be able to purchase additional premium requests beyond their monthly allowance, with administrators able to manage these settings.