Skip to main content
Moneyline
All guides
Beginner6 minAgents

Connect the MCP server to an agent

Give a model tool access to your documents, with writes locked off and untrusted content labelled.

The MCP server exposes Moneyline as tools any MCP client can call. It ships in the box and runs over stdio, so there is nothing to deploy.

Configure the client

claude_desktop_config.jsonjson
{
  "mcpServers": {
    "moneyline": {
      "command": "npx",
      "args": ["-y", "@moneyline/mcp"],
      "env": { "MONEYLINE_API_KEY": "sk_live_..." }
    }
  }
}

Add MONEYLINE_BASE_URL to point at a self-hosted instance. The tool surface is identical, and documents never leave your network.

Use a read-only key

Issue a key scoped to reads for anything exploratory. The four tools that write are gated behind an explicit scope, so a read key makes an entire class of mistake impossible rather than discouraged.

bashbash
moneyline keys create --name "agent-readonly" --scopes documents:read,analytics:read

Ask for outcomes, not calls

The tools are shaped around intent, so the useful prompt is the question you actually have. "What is this applicant's income and are there fraud flags?" resolves in one call. Asking the agent to parse, then enrich, then score makes it orchestrate a sequence the server already handles.

Verify the connection

bashbash
# List the tools the server advertises, without a client.
npx -y @moneyline/mcp --list-tools

Keep going

Run it yourself

Everything in this guide works against the open-source stack. Clone it, or start on the hosted API.