Setup guide

Connecting OmniMem to GitHub Copilot

GitHub Copilot in VS Code has supported MCP servers natively since VS Code 1.99. It speaks Streamable HTTP with custom headers, which is all OmniMem needs.

Requirements

Quick setup

Create or edit .vscode/mcp.json in your project root:

{
  "servers": {
    "omnimem": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp"
    }
  }
}

Copilot uses "servers" as the top-level key, not "mcpServers". Easy to miss.

If you're running the desktop app, Copy MCP URL in the tray or menu bar menu gives you the address.

If OmniMem has an access token set (MCP_AUTH_TOKEN, or Access token on the Configuration page), use an input variable so VS Code prompts for it and stores it securely:

{
  "servers": {
    "omnimem": {
      "type": "http",
      "url": "http://127.0.0.1:8765/mcp",
      "headers": {
        "Authorization": "Bearer ${input:omnimem_token}"
      }
    }
  },
  "inputs": [
    {
      "type": "promptString",
      "id": "omnimem_token",
      "description": "OmniMem access token",
      "password": true
    }
  ]
}

VS Code asks once and remembers. You can also reference an environment variable directly with ${env:OMNIMEM_TOKEN}.

Coming from 6.x, change "type": "sse" to "type": "http" and /sse to /mcp.

Global vs project config

Scope How to get there Use when
Workspace .vscode/mcp.json in the project root Share the config with your team through version control
User (global) Command palette: MCP: Open User Configuration OmniMem should be available in every project

Don't configure the same server in both, it causes conflicts.

Agent mode is required

MCP tools are only available in Copilot's Agent mode, not Ask or Chat. To use OmniMem:

  1. Open Copilot Chat (Ctrl/Cmd + Shift + I)
  2. Pick Agent from the mode dropdown
  3. The OmniMem tools are there

In Agent mode Copilot can call OmniMem on its own as part of a multi-step task.

Using OmniMem with Copilot

Verifying the connection

VS Code asks you to trust a new MCP server the first time it starts it. Approve it, then ask in Agent mode:

Can you call the OmniMem health tool?

You should get back the record and vector counts per namespace, whether the embedding model is loaded, and the uptime.

Known quirks

Notes

By Ric HarveyUpdated All guidesThis guide on Forgejo