GitHub Copilot CLI should fully integrate and leverage features from VS Code Copilot Chat setup and capabilities

[bartlettroscoe]

Describe the feature or problem you'd like to solve

GitHub Copilot CLI should be a CLI/batch interface for the VS Code Copilot Chat setup if that has already been setup for an existing VS Code project

Proposed solution

At face value, it would look like GitHub Copilot CLI is just another proprietary AI coding agent CLI like Claude Code in calling proprietary external models. If that is the case, this is a disappointment and a missed opportunity.

Note the features that VS Code Copilot Chat currently has or will soon have that require non-trivial setup for a VS Code project:

• Custom prompt templates and management
• Custom MCP servers
• Custom model endpoints
• Calling fully local models
  • https://code.visualstudio.com/docs/copilot/customization/language-models#_bring-your-own-language-model-key
  • Add support for generic OAI endpoints microsoft/vscode-copilot-chat#621 (completed released yet but should be in VS Code Sept version released in October?)
• Using VS Code Copilot Chat with local models in enterprise air-gaped settings:
  • Model providers should be community driven microsoft/vscode#249605 (backlog)
• Codebase context gathering uses the LSP sever and back-end indexing for the languages to more effectively gather context (instead of dumb token-hogging find/grep or rg). See:
  • https://code.visualstudio.com/docs/copilot/reference/workspace-context
  • https://code.visualstudio.com/docs/copilot/chat/copilot-chat-context#_workspace-indexing

For example, the latter issue of context gathering for C++ projects requires a local CMake configuration of the project and the clangd implementation of the LSP to provide this indexing. A dump CLI agent will not have any of this (and will have to fall back on slow and token-hogging find/grep).

Any and all of the features and behaviors provided by a custom setup of VS Code Copilot Chat should be available in a CLI/batch interface as per the VS Code backlog item:

• Add scripting interface/batch mode for Copilot Agent Mode? microsoft/vscode#254473

What is the relationship between VS Code Copilot Chat and GitHub Copilot CLI? Are they completely separate products with no relation? If GitHub Copilot CLI is just going to be a independent AI agent that goes off of an AGETNS.md file like Codex CLI or Claude Code, then what is the advantage using GitHub Copilot CLI over Codex CLI? Codex CLI is open source and I can use it to call locally hosted models with zero external networking (and we are doing that right now with good success). Without leveraging the significant setup of a VS Code project with the VS Code Copilot Chat setup, why would anyone in enterprise that has to use local models bother with GitHub Copilot CLI?

Microsoft is making a huge mistake if they are not leveraging the connection to VS Code Copilot Chat with a CLI/batch AI agent interface. That connection is a huge advantage that other AI coding agent CLIs can't easily match.

What is strategy here?

Example prompts or workflows

No response

Additional context

No response

[bartlettroscoe]

I would go a step further and argue that GitHub Copilot CLI should behave identically to VS Code Copilot Chat in a VS Code project that has been set up properly. We need an CLI/batch interface to test prompts, custom MCP servers, etc. on our own custom benchmark suites, as per:

• Add scripting interface/batch mode for Copilot Agent Mode? microsoft/vscode#254473

[bartlettroscoe]

Related:

• Integration of MCP settings with VS Code #39

[whinc]

vsode copilot and github copilot cli should work together providing unified user experience.

[nathfavour]

seconded. CLI (backend) first development is significantly more easily scaled and maintained.

[brunocasado]

where is the "free" models?

[petrroll]

+1

[thawkins]

Additionally for us enterprise users, all and any copilot admin configuration should be picked up from the existing vsc setup. Including things Ike enabling features and models, and enforcing things like MCP registries.

[tikazyq]

where is the "free" models?

Related to #248. See my comment here: #248 (comment)

[bartlettroscoe]

Any updates on this? Will Copilot CLI pick up all the settings and features from an existing VS Code project like sematic search and LSP search, custom prompts, tasks, etc.? We we really need is a programmatic interface to drive the agent in a VS Code project. That would be huge!

[examon]

We addressed the main asks here across several releases. The CLI now reads MCP server configurations from .vscode/mcp.json, loads instruction files from .github/instructions/, supports custom and local model endpoints via the COPILOT_PROVIDER_* environment variables, and includes an LSP client configurable through .github/lsp.json for language-aware code intelligence. Non-interactive usage is available via -p with JSONL output for automation and CI. The exact VS Code .prompt.md template format is not yet supported, but the CLI offers equivalent customization through instructions, custom agents, and skills. If there is a specific capability you still cannot reproduce in the CLI, please feel free to reopen this issue.