LakeXpress MCP Server

The arpe.io MCP server (arpeio-mcp) gives AI assistants a set of tools for the arpe.io products: FastBCP, FastTransfer, MigratorXpress, LakeXpress and the arpe.io ADBC drivers. MCP (Model Context Protocol) is the open standard that Claude, ChatGPT, Cursor, VS Code and most other assistants use to call external tools.

For LakeXpress, the assistant can ask the server to validate a set of parameters and render the exact LakeXpress command, look up what LakeXpress supports, read the release notes and search this documentation. If LakeXpress is installed on your machine, it can also run the command.

The server runs in one of two modes:

• Command builder, the default. The tools build, validate and explain commands; no LakeXpress binary is needed. The hosted server works this way, so it never connects to your databases.
• Execution. When you install the server on your own machine and point it at your LakeXpress binary, the assistant can also run the command it built, once you confirm it.

MCP server or agent skill?

The MCP server gives the assistant tools to call. The agent skill is a set of instructions that teaches it how the product works. The two work well together, so install both if your assistant supports them.

Connect your assistant

arpe.io hosts the server, so there is nothing to install. Add this URL to your assistant:

https://arpe-io-arpeio-mcp.hf.space/sse

Claude Code

claude mcp add --transport sse arpeio https://arpe-io-arpeio-mcp.hf.space/sse

Claude Desktop

Add the server to your Claude Desktop configuration file:

• Linux: ~/.config/Claude/claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "arpeio": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://arpe-io-arpeio-mcp.hf.space/sse"
      ]
    }
  }
}

ChatGPT

Custom connectors need a paid plan (Plus, Pro, Team or Enterprise).

1. Open ChatGPT in your browser, go to Settings > Apps and connectors.
2. Open Advanced settings and enable Developer mode.
3. Go to Connectors > Browse connectors > Add a new connector.
4. Set the URL to https://arpe-io-arpeio-mcp.hf.space/sse and save.

Cursor

Open Cursor Settings, search for "MCP" and add a server with this configuration:

{
  "mcpServers": {
    "arpeio": {
      "url": "https://arpe-io-arpeio-mcp.hf.space/sse"
    }
  }
}

VS Code (Copilot)

Run MCP: Open User Configuration from the Command Palette and add:

{
  "servers": {
    "arpeio": {
      "url": "https://arpe-io-arpeio-mcp.hf.space/sse",
      "type": "sse"
    }
  }
}

Gemini CLI

Add the server to ~/.gemini/settings.json:

{
  "mcpServers": {
    "arpeio": {
      "uri": "https://arpe-io-arpeio-mcp.hf.space/sse"
    }
  }
}

Windsurf

Add the server to ~/.codeium/windsurf/mcp_config.json (Windows: %USERPROFILE%\.codeium\windsurf\mcp_config.json):

{
  "mcpServers": {
    "arpeio": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://arpe-io-arpeio-mcp.hf.space/sse"
      ]
    }
  }
}

Le Chat (Mistral)

Available on all plans, including the free one.

1. Go to Intelligence > Connectors.
2. Click Add connector > Custom MCP Connector.
3. Set the name to Arpe.io and the URL to https://arpe-io-arpeio-mcp.hf.space/sse.
4. Leave authentication disabled and click Create.

Instructions for HuggingChat and Kiro are in the server README.

Run LakeXpress from the assistant

To let the assistant run LakeXpress, install the server on the machine where LakeXpress is installed (Python 3.10 or later):

pip install arpeio-mcp

Then register the arpeio-mcp command instead of the hosted URL, with the path to your binary. In Claude Code:

claude mcp add --transport stdio --env LAKEXPRESS_PATH=/path/to/LakeXpress --env FASTBCP_DIR_PATH=/path/to/FastBCP/directory arpeio -- arpeio-mcp

In Claude Desktop, and in other clients that use the mcpServers format:

{
  "mcpServers": {
    "arpeio": {
      "command": "arpeio-mcp",
      "env": {
        "LAKEXPRESS_PATH": "/path/to/LakeXpress",
        "FASTBCP_DIR_PATH": "/path/to/FastBCP/directory"
      }
    }
  }
}

Variable	Description
LAKEXPRESS_PATH	Path to the LakeXpress binary. Without it, the server only builds commands.
FASTBCP_DIR_PATH	Directory holding the FastBCP binary that LakeXpress uses for the exports.
LAKEXPRESS_TIMEOUT	Execution timeout, in seconds. Optional.
LAKEXPRESS_LOG_DIR	Directory for the execution logs. Optional.
LOG_LEVEL	Server log level: DEBUG, INFO (default), WARNING or ERROR.

Tools

Tool	What it does
lakexpress_info	Reference information, picked with action: capabilities, workflow tips, version detection.
lakexpress_preview_command	Builds any LakeXpress command (lxdb_*, config_*, sync, run, status, cleanup…) and warns about missing required options.
lakexpress_execute_command	Runs the previewed command. Needs a local LakeXpress binary and confirmation: true.

Tools shared by all arpe.io products:

Tool	What it does
arpe_quick_start	Picks the right arpe.io product for a use case described in plain English and returns a workflow guide.
arpe_get_status	Shows which arpe.io binaries the server can run, and which are in command-builder mode only.
arpe_release_notes	Returns release notes for a product, or for one version.
search_docs	Full-text search over the arpe.io documentation sites, this one included, and the arpe.io blog.

How a request flows

1. You describe what you want, for example: "Set up a LakeXpress sync that exports the sales schema from SQL Server to Parquet on S3 and publishes it to Snowflake".
2. The assistant calls lakexpress_preview_command. The tool checks the parameters and returns the LakeXpress command, an explanation, and warnings when an option does not exist in the LakeXpress version you run.
3. You review the command. Passwords are masked in the displayed version.
4. If execution is enabled, the assistant calls lakexpress_execute_command. The tool refuses to run unless the call carries confirmation: true, and it returns the exit code, the output and a diagnosis of known errors.

Most clients also ask for your approval before each tool call.

Prompts and resources

Clients that support MCP prompts (Claude Desktop, Cursor and others) list conversation starters: lakehouse-pipeline for LakeXpress, and troubleshoot to explain a failed run from its error message. The server also exposes the LakeXpress capability matrix as the MCP resource arpeio://capabilities/lakexpress-capabilities.

Reporting issues

If a tool returns a wrong command or wrong information, open an issue at github.com/arpe-io/arpeio-mcp/issues with the prompt you gave and the tool output (with passwords and licence text removed).