Rillet MCP Server

Rillet is an AI-native ERP whose general ledger acts as the system of record for finance teams. The Rillet MCP server is a remote, provider-hosted endpoint that lets AI clients like Claude Desktop, Claude Code, Cursor, and Windsurf query live Rillet data and read Rillet's API documentation directly. Instead of exporting CSVs or writing custom integrations, finance and engineering teams can ask questions in plain English and get answers grounded in real GL transactions, customers, invoices, and accounts.

The server is hosted by Rillet at https://docs.api.rillet.com/mcp and uses the same Bearer-token API keys as the Rillet REST API, with separate sandbox and production environments. Permissions follow Rillet's existing role-based access controls, so what an AI client can see matches what the authenticated user can see in the product. Rillet also offers a one-click install through the Claude Marketplace for users who prefer not to configure the server via CLI.

Under the hood the MCP exposes the Rillet accounting domain: chart of accounts, customers, invoices, payments, credit memos, bills, vendors, bank transactions, journal entries, and reporting endpoints such as balance sheet and income statement. This makes it useful both for ad-hoc finance Q&A ("what is my MRR?", "any overdue invoices?") and for engineering tasks like generating integration code that reflects a real chart of accounts.

Tools

Prerequisites

• A Rillet account with API access enabled.
• A Rillet API key. Generate one from your Organization Settings page in Rillet. Keys are environment-specific (sandbox vs production).
• An MCP-capable client (Claude Desktop, Claude Code, Cursor, Windsurf, etc.).

Option 1: Claude Marketplace (one-click)

Open the Claude Marketplace, search for Rillet, toggle it on, and authenticate with your Rillet API key. No CLI configuration needed.

Option 2: Claude Code CLI

Register the server for the sandbox environment:

claude mcp add --transport http rillet-mcp-sandbox \
  https://docs.api.rillet.com/mcp \
  --header "Authorization: Bearer YOUR_SANDBOX_KEY"

And for production:

claude mcp add --transport http rillet-mcp-prod \
  https://docs.api.rillet.com/mcp \
  --header "Authorization: Bearer YOUR_PROD_KEY"

Verify the connection:

claude mcp list

Option 3: Claude Desktop / generic MCP client config

Add an HTTP MCP server entry pointing at the Rillet endpoint with your API key as a Bearer token:

{
  "mcpServers": {
    "rillet": {
      "transport": "http",
      "url": "https://docs.api.rillet.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_RILLET_API_KEY"
      }
    }
  }
}

When prompting the agent, you can disambiguate environments by saying things like "Use the Rillet MCP in sandbox."

• Ask Claude for live financial status: "What is my current cash balance?" or "What are total Q1 expenses by category?" without leaving the chat.
• Run AR follow-ups: "List all overdue invoices over 30 days past due and group by customer."
• Spot-check the close: "Have all bank accounts been reconciled for April?" or "Show me journal entries posted yesterday."
• Generate integration code for the Rillet API that references your actual chart of accounts and customer records.
• Power finance Q&A in IDEs (Cursor, Windsurf) for engineers building on top of Rillet without context-switching to docs.

• "Use the Rillet MCP in production. What customers do I have with overdue invoices, and what is the total past-due amount?"
• "Pull last month's income statement from Rillet and summarize the largest expense movements vs the prior month."
• "List all journal entries posted in the last 7 days that touch the deferred revenue account."
• "Generate a TypeScript client snippet for creating an invoice in Rillet, using my real chart of accounts."
• "How many active customers are in our Rillet sandbox, and which have no invoices in the past 90 days?"

• Officially built and hosted by Rillet, no self-hosting or community fork required.
• Live, transactional data: every answer traces back to actual Rillet records rather than a stale export.
• Respects Rillet's role-based access controls, so AI access matches the user's existing permissions.
• Separate sandbox and production endpoints make it safe to experiment before pointing agents at real books.

• Requires a paid Rillet account and API key; not useful unless your finance team already uses Rillet.
• Rate-limited at 60 requests per rolling minute on the underlying API, which can throttle long agent runs.
• The public docs do not enumerate exact MCP tool names, so discovery happens at runtime via the client.

• QuickBooks MCP servers (community-maintained) for SMB accounting data access.
• Stripe MCP server for revenue and payment data, though not a full GL.
• Direct use of the Rillet REST API for deterministic, scripted workflows where natural-language access is not required.