Connect your AI to LedgerBeaver

LedgerBeaver speaks the Model Context Protocol (MCP). If your AI client speaks MCP, you can give it your books in about a minute — your AI, your tokens, your control.

Last updated: June 11, 2026 · Private beta

Connect in 60s Per-client setup Scope & tool reference Security & control

Connect in 60 seconds

You'll need a connection token. Generate one inside the app under Connect AI (app.ledgerbeaver.com/connect). Then point any MCP-compatible client at the LedgerBeaver server:

~/connect.sh

# Add LedgerBeaver to any MCP-compatible client
$ mcp add ledgerbeaver \
    --url     https://mcp.ledgerbeaver.com \
    --token   $LB_TOKEN \
    --scope   ap,ar,gl,approvals

Prefer a config file? Most clients read an mcp.json:

~/.config/agent/mcp.json

{
  "servers": {
    "ledgerbeaver": {
      "url": "https://mcp.ledgerbeaver.com",
      "token": "$LB_TOKEN"
    }
  }
}

Keep your token secret. It carries your scopes. Set it as an environment variable (LB_TOKEN) rather than pasting it into a file you might commit. You can revoke and reissue tokens anytime from the Connect AI screen.

Per-client setup

The connection is the same everywhere — only the file location and field names differ. A few common clients:

Claude (Desktop, Code, Web)

Claude supports MCP natively. Add LedgerBeaver as a remote MCP server in your Claude settings, or drop it into your Claude Code config:

claude config (mcpServers)

{
  "mcpServers": {
    "ledgerbeaver": {
      "url": "https://mcp.ledgerbeaver.com",
      "headers": { "Authorization": "Bearer $LB_TOKEN" }
    }
  }
}

Cursor

Add the server to .cursor/mcp.json in your project (or your global Cursor config). Use the mcp.json shape above.

Codex (OpenAI), Gemini CLI, OpenCode & other MCP clients

Any client that implements the MCP transport can connect with the same URL, token, and scopes. Point it at https://mcp.ledgerbeaver.com and pass your token as a Bearer header. If your client speaks MCP, it works.

Scope & tool reference

LedgerBeaver exposes its capabilities as MCP tools, grouped into four scopes. Request only the scopes your agent needs — tokens are scoped per-connection.

Scope	What your AI can do
`ap` Payables	Read bills from the inbox, extract fields, detect duplicates, run three-way match against POs and receipts, code to the GL, and schedule approved payments.
`ar` Receivables	Create and send customer invoices, track aging, run dunning / follow-up sequences, and record payments (Stripe + ACH).
`gl` General Ledger	Read the chart of accounts, draft journal entries, reconcile bank transactions, and pull reports (P&L, balance sheet, cash flow, trial balance). GL postings are human-approved and enforced at the database.
`approvals`	Surface items awaiting sign-off, route by threshold, and record approvals with reason codes. Your AI proposes; you approve.

Once connected, run a listTools call from your client to see the exact tool names and JSON schemas available to your token — that list is always the source of truth for what's callable.

Security & control

You stay in control. Conservative Mode keeps write actions gated: your AI proposes, a human approves what crosses the threshold. The rest flows.
Scoped tokens. Each connection only gets the scopes you grant, and you can revoke a token at any time.
We never see your model traffic. Your AI runs on your provider (Anthropic, OpenAI, whoever). LedgerBeaver only sees the tool calls and their results.
Tamper-evident audit trail. AR, AP, and GL records are SHA-256 hash-chained, so every action is recorded and verifiable.

Questions? support@ledgerbeaver.com.