Skip to main content

WrenAI MCP

Overview​

The WrenAI MCP connects your AI agent to WrenAI so you can query your data, generate SQL, build charts, and get business insights using natural language — right from the tools you already use.

How It Works​

Once connected, your AI agent can talk to your WrenAI project through a set of built-in actions: asking end-to-end questions, generating and running SQL, creating charts, and summarizing results. You install it once per platform (claude.ai, Claude Desktop, or Claude Code), and the companion analyst skill teaches your agent how to use it well.

Before You Begin​

Before installing, make sure the following prerequisites are in place:

  1. Enable the MCP toggle in your WrenAI project. Go to Project settings → MCP connection and turn on the toggle to allow external AI clients to connect via MCP.

    Enable MCP connection toggle

  2. Use a company (organization) account with custom connector permission enabled. If you're on Claude's Team/Enterprise plan, your workspace admin must enable the Add custom connectors permission for your account. Without it, you won't see the "Add custom connector" option in Claude, and you also won't be able to authorize the app on first connect.

    Authorize app on first connect

Install by Platform​

1. claude.ai (Web) and Claude Desktop (chat/cowork/code)​

Add MCP connector + create skill. Auth is handled via OAuth.

Step 1 — Add MCP Connector:

  1. Go to Customize → Connectors → Add custom connector

    Add custom connector menu

  2. Give a name to your mcp

  3. URL: Copy your project MCP url in wrenai.

  4. Auth: OAuth (you'll be redirected to log in with your WrenAI account)

    Add custom connector dialog

  5. Log in with your WrenAI account.

    Log in to WrenAI

  6. Authorize the app.

    Authorize App

Step 2 — Upload Skill:

  1. Open Customize → Skills → Create a skill → Upload a skill
  2. Upload web/claude/wrenai-analyst/SKILL.md
tip

Generate the skill files locally:

git clone https://github.com/canner/wrenai-mcp.git && cd wrenai-mcp && ./setup --host web

2. Claude Code (CLI)​

note

Claude Code's HTTP transport auto-discovers OAuth and hangs if the server advertises /.well-known/oauth-authorization-server. The stdio proxy bypasses this.

# Step 1: Clone the repo
git clone https://github.com/canner/wrenai-mcp.git
cd wrenai-mcp

# Step 2: Add MCP server (stdio proxy with API key)
#   Replace {node_path} with your Node.js >= 20 binary (e.g. ~/.nvm/versions/node/v20.20.1/bin/node)
#   Replace {your_project_id} and {your_api_key} with your WrenAI credentials
cat > .mcp.json << 'EOF'
{
  "mcpServers": {
    "wrenai": {
      "type": "stdio",
      "command": "{node_path}",
      "args": [
        "{absolute_path_to_repo}/mcp-proxy.mjs",
        "https://cloud.getwren.ai/api/mcp/{your_project_id}",
        "{your_api_key}"
      ]
    }
  }
}
EOF

# Step 3: Restart Claude Code

# Step 4 (optional): Install the data analyst skill
/install canner/wrenai-mcp

Generate an API key in your WrenAI project settings.

Done. Try /wrenai-analyst tell me about my data.

Row-Level Security Compatibility​

Currently, the WrenAI MCP does not support Row-Level Security (RLS). Support for RLS is planned for a future release.

Available Skills​

  • setup-wrenai-mcp — One-time setup that connects your agent to a WrenAI instance.
  • wrenai-analyst — Data analyst skill for SQL, charts, and insights from natural language.