Skip to content

IDEs and editors

Seven editors with a built-in agent. For agents that arrive as a VS Code extension — Cline, Roo Code, Kilo Code, Continue — see VS Code extensions instead; they keep their own config files, separate from VS Code's.

Throughout: replace "command": "uvx", "args": ["localmem-mcp"] with "command": "localmem-mcp" and no args if you installed with pip.

Cursor

Create .cursor/mcp.json in the project, or ~/.cursor/mcp.json to have it everywhere:

{
  "mcpServers": {
    "localmem": {
      "command": "uvx",
      "args": ["localmem-mcp"]
    }
  }
}

Cursor picks it up without a restart. Check Settings → MCP for a green dot and the four tools.

MCP tools are only called in Agent mode, not in Ask. To make Cursor reach for memory on its own, add a rule in .cursor/rules/memory.mdc:

---
alwaysApply: true
---
Search localmem with `search_memory` before asking about project context.
Save durable decisions with `store_memory`.

Windsurf

Edit ~/.codeium/windsurf/mcp_config.json:

{
  "mcpServers": {
    "localmem": {
      "command": "uvx",
      "args": ["localmem-mcp"]
    }
  }
}

Then Settings → Cascade → MCP Servers → Refresh. Windsurf shows a hammer icon in Cascade with the tool count once the server is live.

Project instructions go in .windsurfrules.

Zed

Root key is context_servers, and args is required even when empty — a stdio entry without it silently fails to load.

Open settings with Cmd+Shift+P → zed: open settings, or edit ~/.config/zed/settings.json (%APPDATA%\Zed\settings.json on Windows):

{
  "context_servers": {
    "localmem": {
      "command": "uvx",
      "args": ["localmem-mcp"],
      "env": {}
    }
  }
}

You can also add it through the UI: Settings → AI → MCP Servers → Add Server → Add Local Server, which writes the same JSON.

Zed prompts before every tool call by default. To auto-approve just the reads:

{
  "tool_permissions": {
    "mcp:localmem:search_memory": "allow",
    "mcp:localmem:recall_memory": "allow",
    "mcp:localmem:memory_stats": "allow",
    "mcp:localmem:store_memory": "confirm"
  }
}

data did not match any variant of untagged enum ContextServerSettingsContent

Zed's log for a malformed entry. Almost always a missing args, or a nested "command": { "path": ..., "args": ... } object instead of a flat "command": "uvx" string.

VS Code (Copilot agent mode)

The root key is servers, not mcpServers. Pasting a Cursor or Claude Desktop config here does nothing at all, with no error.

Fastest path:

code --add-mcp '{"name":"localmem","command":"uvx","args":["localmem-mcp"]}'

Or create .vscode/mcp.json in the workspace:

{
  "servers": {
    "localmem": {
      "type": "stdio",
      "command": "uvx",
      "args": ["localmem-mcp"]
    }
  }
}

For every workspace, run MCP: Open User Configuration from the Command Palette and add the same block there.

Two more things:

  • Agent mode is required. MCP tools are invisible in Ask and Edit mode. Open Copilot Chat and pick Agent from the mode dropdown.
  • Verify with MCP: List Servers from the Command Palette.

Project instructions go in .github/copilot-instructions.md.

Dev Containers

Put the server in devcontainer.json and it starts with the container:

{
  "customizations": {
    "vscode": {
      "mcp": {
        "servers": {
          "localmem": { "command": "uvx", "args": ["localmem-mcp"] }
        }
      }
    }
  }
}

Point LOCALMEM_DB_PATH at a mounted volume if you want memories to survive the container being rebuilt.

JetBrains AI Assistant

Works across IntelliJ IDEA, PyCharm, WebStorm, GoLand, and the rest.

  1. Press Ctrl+Alt+S (Cmd+, on macOS) to open Settings.
  2. Tools → AI Assistant → Model Context Protocol (MCP).
  3. Click Add, choose the stdio connection type, and paste:
{
  "mcpServers": {
    "localmem": {
      "command": "uvx",
      "args": ["localmem-mcp"]
    }
  }
}

You can also reach this screen by typing / in the AI chat and choosing Add Command.

If you use Junie inside the same IDE, it reads its own config — see Junie.

GUI launch and PATH

A JetBrains IDE started from Spotlight or the Dock doesn't get your shell's PATH. If the server won't start, use the absolute path from which uvx.

Trae

Two ways in, both under Settings → MCP:

  • Add → Manually add, then paste the JSON below.
  • Or click Raw config (JSON) and edit mcp.json directly.
{
  "mcpServers": {
    "localmem": {
      "command": "uvx",
      "args": ["localmem-mcp"]
    }
  }
}

For a project-scoped server, create .trae/mcp.json in the repo root and turn on Enable project-level MCP in Settings → MCP.

Trae's command field must not contain spaces — it's parsed as a single executable, with everything else going in args. If the model download makes startup slow, raise the timeout:

"env": { "START_MCP_TIMEOUT_MS": "60000" }

Google Antigravity IDE

  1. Click … at the top of the agent side panel → MCP Servers.
  2. Manage MCP Servers → View raw config.
  3. Add localmem to mcp_config.json:
{
  "mcpServers": {
    "localmem": {
      "command": "uvx",
      "args": ["localmem-mcp"]
    }
  }
}

The file is ~/.gemini/config/mcp_config.json globally, or .agents/mcp_config.json in the workspace. Save, then hit refresh in the Installed MCP Servers list.

Same two gotchas as the CLI: no type field, and Antigravity doesn't inherit your shell's PATH, so use the absolute path to uvx if the server won't start. See Antigravity CLI for the rest.

Next