MCP Server

Recce exposes its tools to AI agents, enabling integration with tools like Claude Code. This allows you to interact with Recce analysis through natural language queries directly in your Claude Code sessions.

Installation

Install Recce with the MCP extra dependency:

pip install 'recce[mcp]'

Method 1: MCP Server (Stdio)

Configure Recce as an MCP server with stdio transport. Claude Code will automatically launch the MCP server when you start a session.

Configure the MCP server for your dbt project:

cd my-dbt-project/
claude mcp add --scope project recce -- recce mcp-server

Run Claude Code

claude

Verify the connection by running the /mcp command in Claude Code:

> /mcp
╭────────────────────────────────────────────────────────────╮
│ Manage MCP servers                                         │
│                                                            │
│ ❯ 1. recce            ✔ connected · Enter to view details  │

Ask Claude Code about your changes
```
> Understand changes from Recce.
```

Method 2: MCP Server (SSE)

Alternatively, launch a standalone MCP server that Claude Code connects to via HTTP-SSE. This requires running the server in a separate terminal.

Start the standalone MCP server:

cd my-dbt-project/
recce mcp-server --sse

In a separate terminal, configure the MCP server for your dbt project:

cd my-dbt-project/
claude mcp add --transport sse --scope project recce http://localhost:8000/sse

run Claude Code:

claude

Verify the connection by running the /mcp command in Claude Code:

> /mcp
╭────────────────────────────────────────────────────────────╮
│ Manage MCP servers                                         │
│                                                            │
│ ❯ 1. recce            ✔ connected · Enter to view details  │

Ask Claude Code about your changes:
```
> Understand changes from Recce.
```