# vecscope MCP server

> Diagnoses vector search implementations by providing deterministic evidence about embedding space geometry, hygiene, and index performance.

- Repository: https://github.com/076923/vecscope
- Page: https://hostedmcp.org/servers/076923/vecscope
- Transport: stdio
- Stars: 1
- License: MIT
- Latest version: 0.13.0

## Install

### Claude Code

```sh
claude mcp add vecscope -- uvx vecscope
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "vecscope": {
      "command": "uvx",
      "args": [
        "vecscope"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "vecscope": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "vecscope"
      ]
    }
  }
}
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "vecscope": {
      "command": "uvx",
      "args": [
        "vecscope"
      ]
    }
  }
}
```

### Codex (`~/.codex/config.toml`)

```toml
[mcp_servers.vecscope]
command = "uvx"
args = ["vecscope"]
```

### Gemini CLI (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "vecscope": {
      "command": "uvx",
      "args": [
        "vecscope"
      ]
    }
  }
}
```

## FAQ

### Does vecscope measure retrieval relevance?

No, it does not measure relevance. It provides evidence that search is not sabotaging retrieval, but it is not proof that retrieval is good.

### What file formats are supported for embeddings?

vecscope supports .npy, .parquet, .jsonl, .json, and .csv formats.