Month End Close Agent

What it does

Month End Close Agent helps accounting and controller teams run a month-end close review. You provide the ledger (journal entries + reconciliation statuses) and the close period directly — there is no live accounting API and no credentials (connector-free). It runs a multi-step loop: normalize the data, build a close checklist with per-step status, detect anomalies, and synthesize an advisory review summary. The checklist and anomaly detection are fully deterministic and code-derived: it flags unbalanced entries, outlier amounts, duplicates, missing categories, period mismatches, and unreconciled accounts. An LLM is used ONLY for the advisory summary narrating those findings, with a deterministic template fallback when no model is available. It is strictly read-only: it never posts entries, reconciles, closes a period, or moves money — the close decision stays with a human. Ledger free-text is treated as untrusted data and sanitized against prompt injection. Tools: - run_full — end-to-end: normalize, build the close checklist, detect anomalies, synthesize an advisory review summary (LLM with template fallback), return the full result JSON. - build_checklist — deterministic close checklist only (step + status), no LLM. - detect_anomalies — deterministic anomaly scan only (unbalanced entries, outliers, duplicates, missing categories, period mismatches, unreconciled accounts), no LLM. - list_capabilities — static capabilities: tools and the anomaly checks performed. - plan_inputs — interview helper returning questions, JSON schema, and a ready-to-edit example for a tool. Provide entries as [{ id, date, description?, account?, category?, debit, credit }]; optionally reconciliations and a checklistTemplate. Array/object arguments may be passed as JSON or a JSON string. Missing required inputs return a structured needs_input payload (questions + schema + example) instead of erroring, and upgrade to a native elicitation form on clients that support it.

Example prompts

• Run run_full on this May ledger for Acme Inc. — build the close checklist, flag anomalies, and give me an advisory summary.
• Run detect_anomalies on these journal entries — unbalanced entries, outliers, duplicates, and unreconciled accounts only.
• Build the month-end close checklist for period 2026-05 from these entries and reconciliations.
• Which accounts are still unreconciled and what's blocking sign-off this period?
• Use plan_inputs for run_full and show me the questions, schema, and a ready-to-edit example.

Tools (5)

Tools the agent exposes — your AI client calls them automatically when it needs them.

• run_full — Run the month-end close agent end-to-end on a caller-provided ledger: normalize the data, build a close checklist, detect anomalies, synthesize an advisory review summary (LLM with template fallback), and return the full result JSON. Read-only; does not post entries or close the period.
• build_checklist — Deterministically build only the month-end close checklist (step + status) from a caller-provided ledger. No LLM.
• detect_anomalies — Deterministically scan a caller-provided ledger for anomalies (unbalanced entries, outlier amounts, duplicates, missing categories, period mismatches, unreconciled accounts). No LLM.
• list_capabilities — List the agent's static capabilities: available tools and the close-checklist + anomaly checks it performs.
• plan_inputs — Plan/brainstorm the inputs for a tool: returns the questions, schema and a ready-to-edit example.