# Loom MCP server

> Acts as an MCP proxy that intercepts oversized tool results and converts them into queryable DuckDB datasets to save model context.

- Repository: https://github.com/buttonmonkey/mcp-loom
- Page: https://hostedmcp.org/servers/buttonmonkey/mcp-loom
- Transport: stdio
- Stars: 1
- License: MIT
- Latest version: 0.3.2

## Install

### Claude Code

```sh
claude mcp add mcp-loom -- npx -y mcp-loom
```

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

```json
{
  "mcpServers": {
    "mcp-loom": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-loom"
      ]
    }
  }
}
```

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

```json
{
  "servers": {
    "mcp-loom": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "mcp-loom"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"mcp-loom","type":"stdio","command":"npx","args":["-y","mcp-loom"]}'
```

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

```json
{
  "mcpServers": {
    "mcp-loom": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-loom"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add mcp-loom -- npx -y mcp-loom
```

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

```toml
[mcp_servers.mcp-loom]
command = "npx"
args = ["-y", "mcp-loom"]
```

### Gemini CLI

```sh
gemini mcp add -s user mcp-loom npx -- -y mcp-loom
```

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

```json
{
  "mcpServers": {
    "mcp-loom": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-loom"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `GITHUB_TOKEN` | no | no |  |
| `LOOM_CONFIG` | no | no |  |

## FAQ

### What happens when a tool result is too large?

Loom intercepts the large tabular result, ingests it into an embedded DuckDB, and provides the model with a compact envelope containing a dataset reference, schema, and sample rows.

### Can I perform joins across data from different servers?

Yes, Loom enables the model to perform joins across datasets originating from different downstream MCP servers.

### How do I configure Loom for my MCP client?

You can point your MCP client to the command `npx -y mcp-loom --config <absolute path to loom.config.json>` using a JSON configuration file.

### Does Loom add data that the original server didn't provide?

No, Loom faithfully flattens what the server sends but cannot add fields that were not present in the original response.
