neuralmind.uk / Use case: VS Code native code intelligence Edit this page on GitHub ↗

Use case: VS Code native code intelligence

Who this is for: VS Code users who want NeuralMind’s semantic search, graph view, and auto-build available without leaving their editor or wiring up a tasks file.

What it unlocks: The NeuralMind VS Code extension turns the CLI into native editor affordances — status bar freshness indicator, command palette queries, an embedded graph panel, and an opt-in hover provider — all talking to the same local Python process, no cloud required.

Installation

# 1. Install NeuralMind (if not already)
pip install neuralmind

# 2. Build your project's index
cd /path/to/your/project
neuralmind build .

# 3. Install the extension
cd /path/to/neuralmind-repo/editors/vscode
npm install && npm run compile

# 4. In VS Code: Extensions → "Install from VSIX" → select editors/vscode/
# Or open editors/vscode/ as a workspace and press F5 to launch a debug session.

What you see

Status bar

The status bar item on the left shows index freshness at a glance:

State Display Meaning
Building ⟳ NeuralMind (spin) Polling on activation
Built, fresh ✓ NeuralMind · 2.1k nodes (green) Index within threshold
Built, stale ⚠ NeuralMind · 2.1k nodes (yellow) Older than autoBuildThresholdHours
Not built ⊘ NeuralMind (red) No .neuralmind/ directory found

Click the status bar item to open the graph view.

Command palette (Ctrl+Shift+P)

Command What it does
NeuralMind: Query Prompts for a question, shows optimised context in the Output panel
NeuralMind: Wakeup Shows the ~600-token project overview in the Output panel
NeuralMind: Skeleton Shows a compact skeleton of the active file
NeuralMind: Build Index Runs neuralmind build with a progress notification
NeuralMind: Probe Runs neuralmind probe and shows recall@k/MRR/answerability
NeuralMind: Open Graph View Opens the force-directed graph panel
NeuralMind: Setup Cline Integration Registers MCP with Cline (restart Cline after)
NeuralMind: Setup VS Code MCP Integration Registers MCP in settings.json (VS Code 1.99+)

Graph panel

NeuralMind: Open Graph View opens the same force-directed graph as neuralmind serve in a VS Code WebviewPanel — force layout, community colouring, semantic search bar, live synapse events via SSE. The panel retains its layout state when hidden.

Workflow

Day-to-day query loop

  1. Open your workspace. The status bar shows current index state.
  2. Press Ctrl+Shift+P → NeuralMind: Query → type your question.
  3. Context appears in the NeuralMind Output panel (optimised, 12-50× smaller than pasting files).
  4. Copy the relevant context into your AI assistant prompt, or let the MCP server inject it automatically.

Auto-build on stale index

On workspace open, if the index is older than neuralmind.autoBuildThresholdHours (default: 24h), the extension shows a notification:

NeuralMind: Index is stale or not built. Build now? [Build]

Click Build to run neuralmind build . in the background.

Hover cards (opt-in)

Enable in Settings:

{
  "neuralmind.enableHover": true
}

Hover over any function or class definition in Python, TypeScript, Go, Rust, Java, or C/C++ to see a compact skeleton of that file’s structure — same output as neuralmind skeleton. Results are cached per-file for 60 seconds. Hover cards only appear when the index is built (checked via the status bar).

MCP integration for VS Code 1.99+

VS Code 1.99 added native MCP client support. Register NeuralMind in one command:

NeuralMind: Setup VS Code MCP Integration

This writes the neuralmind-mcp entry to your VS Code settings.json under mcp.servers. Reload the window to activate. You can also run this from the CLI:

neuralmind install-mcp . --client vscode

Troubleshooting

Status bar stays red / ⊘ NeuralMind

Run neuralmind build . in the terminal. The extension polls neuralmind stats --json every 60 seconds, so it will update automatically after the build completes.

“NeuralMind: CLI not found” notification

Set neuralmind.pythonPath to your virtualenv’s Python:

{
  "neuralmind.pythonPath": "/path/to/venv/bin/python"
}

Graph panel shows “server not running”

The extension starts neuralmind serve on a random port at activation. If the server didn’t start (e.g. no index built yet), build the index first, then run NeuralMind: Open Graph View again.

Hover cards are slow

Hover cards spawn a subprocess per file. For large projects, the first call per file takes ~0.5–2 seconds (cached for 60s after). If that’s too slow, disable hover:

{
  "neuralmind.enableHover": false
}