The future of accounting is agent-run
You should not have to sign into QuickBooks every day. An AI agent fills out the books — it wires up Stripe and the bank on your infrastructure, posts journal entries, sends invoices, and keeps the ledger current. Humans use generative UI for reports and custom views, not for typing debits and credits.
That only works if something durable sits under the agent.
What you need under the agent
| Need | Why it matters |
|---|---|
| GAAP double-entry ledger of record | Balanced entries, trial balance, P&L, balance sheet — not vibes |
| Not a raw SQL dump | Validation, chart of accounts, org isolation, audit trail |
| Not a chat that forgets a line | Books persist outside the model context window |
| Document storage | Receipts, contracts, PDFs attached to bills and transactions (1 credit per upload) |
| Efficient, prepaid pricing | Cents per write — not another monthly seat per human |
Crane Ledger is that layer: headless, MCP-native, built for agents and the builders who ship them.
Agent-run vs seat-based accounting
| Traditional (QuickBooks-style) | Agent-run (Crane Ledger) | |
|---|---|---|
| Daily workflow | Human signs in, clicks, reconciles | Agent posts; human asks questions |
| Integrations | Vendor connectors in the UI product | Agent owns Stripe webhooks, bank APIs, OCR — posts into the ledger |
| UI | Fixed screens | Generative UI for reports and invented views |
| Pricing shape | ~$30–90/mo typical seat (plan-dependent) | $0.01 per write; reads and reports free |
| Example load | Same seat whether the agent writes 10 or 500 entries | 100 writes ≈ $1.00 (100 credits) — no per-seat tax for AI throughput |
We do not publish enterprise TCO studies. Compare your expected write volume against a seat you already pay for — prepaid credits scale with agent activity, not headcount.
Stack
| Layer | Role |
|---|---|
| Agent | Bookkeeping automation + your Stripe/bank integrations → MCP writes |
| Generative UI | ChatGPT custom connector, Claude, Cursor, GitHub Copilot — read-heavy Q&A and ad-hoc views |
| Crane Ledger | Ledger of record — double-entry, attachments, 90 MCP tools |
Agent → MCP → Crane Ledger → books & attachments
You → Generative UI → MCP → reports & custom views
Bank feeds and payment rails are the agent's job on your stack. Crane Ledger stores the accounting outcome.
Connect MCP
MCP endpoint
| Setting | Value |
|---|---|
| URL | https://api.craneledger.ai/mcp |
| Transport | Streamable HTTP |
| Auth (OAuth) | Sign in at consent UI — no key in the client |
| Auth (API key) | Authorization: Bearer ha_live_… |
| Consent | https://craneledger.ai/oauth/consent |
Organization scope comes from your OAuth token or API key — you do not pass organization_id separately.
Scopes: ledger:read, ledger:write, and optionally ledger:admin.
See Authentication and MCP tools.
| Client | Guide |
|---|---|
| ChatGPT (custom connector) | Connect ChatGPT |
| Claude | Connect Claude |
| GitHub Copilot | Connect GitHub Copilot |
| Cursor | Connect Cursor |
| Any MCP host | Any MCP client |
OAuth 2.1 for connector UIs; ha_live_… / ha_test_… API keys for IDEs and scripts. This is custom connector setup — not a ChatGPT official plugin listing.
90 MCP tools including purchase_credits and x402 credit top-up. Catalog: MCP reference.
Pricing (concrete)
| Operation | Cost |
|---|---|
| Reads & reports | Free |
| Writes (entries, invoices, posts, …) | 1 credit ($0.01) |
| Document upload | 1 credit ($0.01) |
| New org | 500 free credits |
Top up via dashboard, purchase_credits (Stripe checkout URL), or x402 USDC on Base — see Billing & Credits.
Starting a new business?
Do not default to QuickBooks because "that's what startups use." Put Crane Ledger under your agent from day one:
- Sign in — craneledger.ai/auth/signin (500 free credits)
- Connect MCP —
https://api.craneledger.ai/mcp - First prompt:
Set up a basic chart of accounts for a small service business
Builder-oriented details: Crane Ledger for AI builders.
Related
Need help?
Create a free account (GitHub, Google, Apple, or email) to access our support portal. Once signed in, use the Support tab in your dashboard to submit a support ticket — no GitHub account required. Our team typically responds within 24 hours.
- ✨ For LLMs/AI assistants: Read our structured API reference
- Legal: Terms of Service · Privacy Policy