# Visual Studio and MCP Setup

This guide covers two related workflows:

1. Running Cortex integration tests from Visual Studio Test Explorer.
2. Connecting an MCP-capable Copilot client to the Cortex MCP server.

## Visual Studio Test Explorer

Visual Studio can run the integration tests because they are ordinary xUnit tests.

### Steps

1. Open `Cortex.slnx`.
2. Start the isolated test stack:

   ```powershell
   .\scripts\deploy-test.ps1
   ```

3. In Visual Studio, open Test Explorer.
4. Run tests under `Cortex.IntegrationTests`.

The tests connect to:

```text
http://localhost:5217
```

Override that by setting:

```powershell
$env:CORTEX_TEST_BASE_URL = "http://localhost:5217"
```

## Connecting a Copilot MCP Client

Cortex uses Streamable HTTP MCP.

The Blazor web UI is not needed for this. It is only a browser-based helper for local testing and memory inspection.

Dev endpoint:

```text
http://localhost:5117/mcp
```

Test endpoint:

```text
http://localhost:5217/mcp
```

## Visual Studio Full IDE

As of this documentation pass, Microsoft has clear public documentation for MCP server configuration in Visual Studio Code. I did not find equivalent official full Visual Studio IDE documentation for a stable `mcp.json` location.

If your Visual Studio installation exposes MCP server configuration through GitHub Copilot or an agent customization UI, use the same Streamable HTTP server details:

```json
{
  "servers": {
    "cortex": {
      "type": "http",
      "url": "http://localhost:5117/mcp"
    }
  }
}
```

If Visual Studio asks for a URL directly, enter:

```text
http://localhost:5117/mcp
```

If your Visual Studio build does not expose MCP server configuration, use Visual Studio for normal coding and Test Explorer, and use Visual Studio Code or another MCP-capable client for Cortex tool calls.

## Visual Studio Code MCP Setup

VS Code supports workspace-level MCP configuration through `.vscode/mcp.json`.

Create:

```text
.vscode/mcp.json
```

Add:

```json
{
  "servers": {
    "cortex": {
      "type": "http",
      "url": "http://localhost:5117/mcp"
    }
  }
}
```

The same example is available at `docs/examples/vscode-mcp.dev.json`.

Then:

1. Start Cortex:

   ```powershell
   .\scripts\deploy-all.ps1
   ```

2. Open the VS Code command palette.
3. Run `MCP: List Servers`.
4. Start or enable `cortex`.
5. Open Chat and select/configure tools.

For the test instance:

```json
{
  "servers": {
    "cortex-test": {
      "type": "http",
      "url": "http://localhost:5217/mcp"
    }
  }
}
```

The same example is available at `docs/examples/vscode-mcp.test.json`.

## Example Prompts

After connecting an MCP client:

```text
Create a Cortex project named Home.
```

```text
Remember in the Home project as a location: my USB-C charging brick is in the blue backpack.
```

```text
Find Home memories about the charging brick.
```

```text
Add a todo in the Home project: buy a replacement USB-C cable.
```

## Trust and Safety

Cortex is local and currently unauthenticated. Treat the endpoint as trusted-local only:

- Keep it bound to localhost for development.
- Do not expose it to the public internet.
- Review MCP tool calls before approving them in a client.
- Use explicit projects to avoid accidental memory mixing.

## References

- [VS Code MCP server configuration](https://code.visualstudio.com/docs/agent-customization/mcp-servers)
- [MCP C# SDK transport documentation](https://csharp.sdk.modelcontextprotocol.io/concepts/transports/transports.html)