Skip to content

GitHub Copilot (VS Code)

Use HydraMem as a memory tool for GitHub Copilot agent mode in VS Code via MCP.

Prerequisites

  • HydraMem on $PATH (hydramem --version).
  • VS Code with GitHub Copilot (agent mode / tools enabled).

Connect (MCP)

Create .vscode/mcp.json in your workspace:

{
  "servers": {
    "hydramem": {
      "type": "stdio",
      "command": "hydramem",
      "args": ["serve", "--transport", "stdio"]
    }
  }
}
User-level (all workspaces)

Add the same servers entry to your VS Code user mcp.json (Command Palette → MCP: Open User Configuration).

HTTP transport (shared server)

If you run HydraMem over HTTP (hydramem serve --transport http), point VS Code at it:

{
  "servers": {
    "hydramem": { "type": "http", "url": "http://localhost:3000/mcp" }
  }
}

Open the Copilot Chat Agent mode, click the tools picker, and enable the hydramem tools.

Built-in HydraMem skills for Copilot

This repository includes six Copilot skills for common memory workflows:

  • hydramem-ingest
  • hydramem-ingest-smart
  • hydramem-query
  • hydramem-reason
  • hydramem-link
  • hydramem-garden

See the full guide for when to use each one and how they handle context injection/retrieval: GitHub Copilot Agent Skills.

Use it

Ask Copilot agent mode things like "search our HydraMem memory for the auth decision" or "remember that we switched to Postgres". It will call hydra_search_tool / remember against your local palace.

Verify

  • VS Code → MCP: List Servershydramem is running.
  • Terminal: hydramem search "…" · hydramem stats --last-7d.

Troubleshooting

  • Server not starting → check hydramem --version resolves; use the absolute path to the console script in command if VS Code's PATH differs.
  • No tools shown → ensure agent mode is on and the server is enabled in the tools picker.

See also