Complementary tools
Claude (Anthropic) is one of the most capable AI assistants available — and Ledger Layer is powered by Claude. Claude powers the AI extraction layer that reads your Excel workbooks, interprets lease terms, and structures unstructured data. But like all language models, Claude cannot produce deterministic, audit-grade accounting output on its own. Ledger Layer is the control layer that makes Claude safe for accounting.
Ledger Layer is powered by Claude and OpenAI for AI extraction. All accounting computation is performed by Ledger Layer's deterministic engine. The invariant is enforced architecturally: AI has zero access to compute functions.
The question is not "Claude or Ledger Layer?" — it's "how do they work together?" Ledger Layer was designed from day one to use AI for what it's good at (reading, interpreting, summarising) and a deterministic engine for what AI cannot do (computing, enforcing, auditing).
Reads your Excel workbook, extracts lease terms, answers questions, drafts disclosure language, queries portfolio via MCP
Computes PV, amortisation schedules, journal entries — deterministically, hash-verified, version-pinned
Audit-ready journals, disclosure packs, structured journal exports (CSV/XLSX/JSON) ready for any ERP or iPaaS, and financial statement impact reports
Ledger Layer is powered by Claude via API for the AI extraction layer. The architectural invariant guarantees that AI has zero access to compute functions — Claude reads and interprets, but never calculates a present value or generates a journal entry. You can also connect Claude to Ledger Layer via MCP — 75+ schema-validated tools give Claude safe, structured access to your accounting data without ever letting it touch the computation layer.
When you connect Claude to Ledger Layer via MCP, Claude gets access to your full accounting data through structured, role-gated tools. Here's what that looks like in practice:
Claude calls portfolio_summary, reads the structured response, and presents a clear breakdown by entity, currency, and standard — with numbers that come directly from the engine.
Claude calls lease_extract_file, processes the AI extraction, and presents the extracted terms for your review. You confirm, and the engine runs. No manual data entry.
Claude calls disclosures_get, reads the engine-verified maturity analysis, and drafts narrative disclosure language. The numbers are Ledger Layer-verified. The words are Claude-drafted.
Claude calls portfolio_analytics, identifies leases approaching renewal or expiry, flags missing IBR rates, and surfaces modification events — all from structured, current data.
Claude queries multiple tools — portfolio_search, journal_search, system_health — and compiles a pre-close checklist with outstanding items, missing data, and recommended actions.
Claude calls lease_detail for a specific lease, reads the modification history, and explains what changed, when, and how it affected the schedule — in plain language.
Claude gets 50+ accounting tools with role-gated access. Your data stays controlled. Auditors get the trail they require. Setup takes under 5 minutes.