Skip to content

feat: expose ShortKit-ML as a GitHub Copilot extension #3

@sebasmos

Description

@sebasmos

Summary

Add GitHub Copilot support so developers can run ShortKit-ML shortcut detection directly from the GitHub Copilot chat panel in Visual Studio Code — the same way the MCP server already enables this for Claude Code and Claude Desktop.

Background

We already expose all 19 detection methods as MCP tools (docs). GitHub Copilot Extensions use a different protocol (the Copilot Extensions API) but the goal is the same: let an AI assistant call run_detector, compare_methods, generate_synthetic_data, etc. from the VS Code Copilot chat sidebar.

Target environment

  • Editor: Visual Studio Code
  • AI assistant: GitHub Copilot (chat panel, @shortkit-ml mention)
  • Protocol: Copilot Extensions API (GitHub App + skillset or agent)

What needs to be done

  • Research — determine whether a Copilot Extension is best implemented as a Copilot Agent (GitHub App + webhook) or a skillset (simpler, function-calling only)
  • Expose the same tool surface as the MCP server:
    • list_methods
    • generate_synthetic_data
    • run_detector
    • get_summary
    • get_method_detail
    • compare_methods
    • run_benchmark
    • generate_report
  • Register the GitHub App and configure the extension manifest
  • Test end-to-end in VS Code — open Copilot chat, type @shortkit-ml run detector on my embeddings, verify tool calls and results render correctly
  • Write automated tests — mirror the existing tests/test_mcp_server.py structure
  • Document — add a Copilot / VS Code section to the MCP server docs page covering installation and example prompts
  • Update README MCP section to mention Copilot alongside Claude

Reference

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions