# Jarca Ledger MCP Interface Contract - Baseline Mode: observed - Baseline Seed: `.stage-pilot/bootstrap/baseline.yaml` - Status: governed - Owner: Jarca / project maintainer - Last Updated (KST): 2026-07-01 - Source Discovery / Batch: bootstrap-baseline, dcy-007, BAT-007, BAT-008, BAT-009 ## Project Summary - Jarca 가계부 전용 MCP 서버와 Streamlit 대시보드로 계좌/카드 거래, 예산, 분류 규칙, import, 현금흐름 분석을 SQLite에 저장하고 운영한다. ## Purpose - This document captures the current external and operator-facing interface baseline. - It is a reference for future Discovery, REQ, Batch, and Release work that changes tool signatures, dashboard behavior, or file import contracts. ## Interface: MCP Tool Surface - Type: internal-service - Actors: Hermes Agent and MCP clients - Entrypoint: `jarca-ledger-mcp` script -> `jarca_ledger_mcp.server:main` - Stability: internal but should preserve existing success response shapes for compatibility. ### Representative Tool Groups - Health: `ledger_health_check`, `get_ledger_db_status` - Operational backup: `create_ledger_backup`, `list_ledger_backups`, `get_ledger_backup_status` (restore is intentionally not exposed as MCP) - Accounts: `create_ledger_account`, `list_ledger_accounts`, `update_ledger_account`; account payloads include additive `liquidity_bucket` for asset-liquidity reporting. - Transactions: `create_ledger_transaction`, `list_ledger_transactions`, `update_ledger_transaction`, `delete_ledger_transaction` - Budgets/reports: `create_ledger_budget`, `list_ledger_budgets`, monthly/category/fixed/budget/investable reports, `get_asset_liquidity_report` - Asset valuation reads/writes: `list_manual_assets`, `create_manual_asset`, `update_manual_asset`, `deactivate_manual_asset`, `preview_asset_snapshot_close`, `close_asset_snapshot`, `get_asset_snapshot_report`, `get_net_worth_timeseries`; these use the same valuation policy as the dashboard and preserve append-only snapshot correction semantics. - Imports: KB Bank PDF, Toss Bank PDF, Toss Securities PDF, SeoulPay XLSX, CSV preview/import where implemented - Categories/rules: category CRUD, rule CRUD, categorization preview/apply, uncategorized summaries - Reconciliation: cross-account preview/run/review queue, self-transfer candidates/linking, cancel-original candidates/linking - Manual balance reconciliation: `get_balance_reconciliation_report` returns preview-only account reconciliation for `manual_only`/`partial_import` accounts. ### Inputs - Structured MCP tool arguments as declared in `server.py` signatures. - File paths and optional passwords for encrypted or password-protected documents. - Transaction/account/category identifiers for update, delete, and link operations. ### Outputs - Python dictionary payloads containing success data, counts, records, previews, reports, or error details. - Backup/status tools return metadata-only DB evidence: path, size, mtime, sha256, integrity check, table row counts, and aggregate totals. - Existing success shape compatibility is preferred; new fields should be additive where possible. - Contract version metadata is not exposed in runtime success payloads for version `1.0`; version governance is tracked in this document and contract regression snapshot metadata. ### Errors - Validation errors for invalid parameters or missing IDs. - Parser errors for unreadable or unsupported source files. - Password/validation errors for encrypted import files. - Candidate-not-found errors for link operations when the chosen pair is not in the preview set. ### Compatibility Rules - Preserve existing success response shapes unless a migration/REQ explicitly approves a breaking change. - MCP success response contract version uses `major.minor`; the initial governed contract version is `1.0`. - Increment `major` for breaking changes, including required key removal, existing key type change, or nested object/list structure changes that break existing consumers. - Increment `minor` for backward-compatible additive changes, including optional field additions or documentation/test metadata that preserves existing payload semantics. - Ambiguous response-shape changes must identify contract impact in the owning REQ/BAT and be reviewed against this document before implementation. - Additive backup/status/list/hash tools are allowed; restore execution must not be added as an MCP tool. - Do not write secrets, raw passwords, raw_payload, full merchant/counterparty dumps, or full account identifiers into persistent output, skill files, or committed documentation. - Source evidence should remain auditable through raw fields and import batch records. ### MCP Response Contract Version Governance - Source of truth: this `docs/interface-contract.md` section defines the active contract version, change classification rules, and runtime exposure policy. - Snapshot metadata: representative contract regression fixtures live under `tests/contract_snapshots/` and must carry the same contract version as this document. - Runtime exposure policy: `contract_version` remains unexposed in MCP success responses for `1.0`; adding it to runtime payloads requires a future approved REQ/migration because it changes every success envelope, even if additive. - Required first-wave regression scopes: health/status, list, report, and mutation success envelope responses. - Additive policy: contract tests intentionally assert a required subset and required types so newly added optional fields do not fail the gate. - Breaking-change policy: removing a required key, changing a required key type, or changing a required nested shape must fail contract regression unless an approved REQ/BAT updates the contract version and snapshot metadata. - Every REQ/BAT that changes MCP success response shapes must state one of: `contract impact: none`, `contract impact: additive minor`, or `contract impact: breaking major`. ### MCP Response Contract Version Log | Version | Date (KST) | Change Type | Source | Notes | | --- | --- | --- | --- | --- | | 1.0 | 2026-06-30 | Initial governed baseline | dcy-007 / REQ-007 / BAT-007 | Formalizes compatibility governance without exposing runtime version fields or changing success payload shapes. | ### Asset Valuation Read/Mutation Tools - `list_manual_assets(active_only=true)` returns `success`, `manual_assets`, `count`, and `valuation_policy`. Each item exposes `name`, `valuation_amount`, `valuation_date`, `liquidity_bucket`, `source_note`, `memo`, and `authority_source='manual_assets'`. - `create_manual_asset`, `update_manual_asset`, and `deactivate_manual_asset` return dcy-007-compatible success/error envelopes. Mutation success includes the affected `manual_asset` payload or deactivation metadata; soft delete (`active=false`) is the default. - `preview_asset_snapshot_close(basis_date='')` returns `success`, `basis_date`, `source='live_authority'`, `is_closed_snapshot=false`, item lists, bucket totals, total assets/liabilities, net worth, and warnings. It is side-effect free. - `close_asset_snapshot(basis_date='', source_note='', memo='', correction_reason='')` writes item-level snapshot rows and returns `source='snapshot'`, `is_closed_snapshot=true`, `snapshot_run_id`, correction metadata, inserted count, totals, and warnings. If rows already exist for the same basis date, `correction_reason` is required and the close appends a correction run instead of overwriting. - `get_asset_snapshot_report(basis_date='')` returns `success`, `basis_date`, `source`, `is_closed_snapshot`, `snapshot_count`, `buckets`, `total_assets`, `total_liabilities`, `net_worth`, `included_items`, `excluded_items`, and `valuation_policy`. - If `asset_snapshots` rows exist for `basis_date`, `source='asset_snapshots'`, `is_closed_snapshot=true`, and values come from the latest stored close run when run metadata exists. - If no snapshot rows exist, `source='live_authority'`, `is_closed_snapshot=false`, and the report reads `accounts.last_balance` plus active `manual_assets`. - `get_net_worth_timeseries(start_date='', end_date='', months=12)` returns `success`, requested date bounds, `points`, `count`, and `valuation_policy`; each point contains `basis_date`, `total_assets`, `total_liabilities`, `net_worth`, `source`, and `item_count`. - Contract impact: additive minor. Existing success payloads are preserved; `get_asset_liquidity_report` gains optional explanatory fields (`manual_assets`, `manual_asset_count`, `included_items`, `excluded_items`, `valuation_policy`) while preserving existing keys, and BAT-009 adds new mutation/preview scopes to contract regression. ## Interface: CI Required-Check Contract - Required-check source of truth in code: `.github/workflows/ci.yml` job names are stable and must be mirrored by repository branch protection settings. - Required-check default set for PR/push merge blocking: `pytest-regression`, `dashboard-app-test`, `dashboard-http-smoke`, `db-safety-smoke`, `contract-regression`, `lint`. - `contract-regression` runs `pytest tests/test_contract_regression.py` against synthetic temp fixtures and representative response snapshot metadata. - `lint` runs `ruff check .` with the conservative BAT-007 initial rule set (`E9`, `F`) as a required blocking gate. - Actual GitHub branch protection toggles are repository settings outside this codebase; this document and the workflow define the intended contract for the operator to apply and audit. - CI contract tests, AppTest, HTTP smoke, and DB safety smoke must use synthetic/temp DBs only and must not read the default operating DB path. ## Interface: Streamlit Dashboard - Type: web-app - Actors: Local operator - Entrypoint: `streamlit run dashboard/app.py --server.address 0.0.0.0 --server.port 8501` - Stability: internal operator UI ### Pages - 요약 - 거래내역 - 확인필요 큐 - 미분류 - 분류 규칙 - 카테고리 - 계좌 - Import ### Inputs - Local SQLite database selected by `JARCA_LEDGER_DB_PATH` or default path. - Uploaded transaction history files. - User edits for categories, rules, transactions, and manual links. ### Outputs - Tables, charts, dialogs, previews, import results, and transaction/category updates. - Browser evidence runs produce local PNG screenshots, DOM snapshot HTML files, and a manifest JSON for QC review. ### UI Contract Rules - Surface `cashflow_kind` where classification affects actual cashflow interpretation. - Keep actual income/expense separated from account movement summaries. - Rule page should support create/edit/delete and condition fields `amount_min`, `amount_max`, `day_of_month`. - Dialog-based triage actions should avoid losing context where possible. - Account page should display/edit `statement_status` and show a minimal manual balance reconciliation report without adding an automatic schedule or dedicated workbench. - Account page should display/edit `liquidity_bucket`; overview uses the shared service valuation report for asset liquidity/composition/debt/net-worth sections. - Overview asset/net-worth display must keep the REQ-005 account-only fallback when `manual_assets`/`asset_snapshots` are empty, and may additionally show manual asset rows when present. - Overview top priority is KST current-month actual income/expense/net cashflow/savings rate based on `transaction_date` and `cashflow_kind`. - UI/dashboard-impacting batches require browser screenshot/snapshot evidence from synthetic/validation DBs. Non-UI batches may omit browser evidence, and docs-only batches do not require it. ## Interface: File Import Contracts - KB Bank PDF: encrypted/password-protected PDF account history; preserve transaction datetime, counterparty, balance, branch, raw summary, and raw payload. - Toss Bank PDF: transaction history PDF parsed into ledger transaction candidates. - Toss Securities PDF: securities transaction history with investment movement, dividend, interest, and external transfer handling. - SeoulPay XLSX: possibly encrypted Office XLSX/OLE file; decrypt with supplied password and import usage details as actual spending evidence. - CSV: generic preview-first import path. ## Shared Constraints - All import flows should support preview before mutation where implemented. - Dedupe should use deterministic source/external IDs or same-account/date/type/amount/merchant evidence. - Import batch metadata should support rollback and audit. - Broad mutation workflows should capture metadata-only DB status and backup evidence before/after risky operations. - Restore execution is limited to script/runbook workflows and explicit approval, not MCP tools. - Manual balance reconciliation is preview/report-first. `REVIEW_NEEDED` suggests manual decision values (`계좌이동`, `실지출`, `실수입`, `투자이동`, `분석제외`, `확인필요`) but does not auto-confirm mutations. - Asset liquidity reporting uses the shared valuation policy: account-backed values from `accounts.last_balance`, non-transaction assets/liabilities from `manual_assets`, `loan` as debt/net-worth only, and missing values excluded with explanation rather than inferred from transaction totals. - Browser evidence screenshots/snapshots must use synthetic or validation DBs only and must not include real account identifiers, source financial filenames, passwords, or raw payloads. - Financial data and passwords are sensitive runtime inputs and should not be committed. ## Current Gaps / Planned Changes - Formal MCP response contract versioning is defined for version `1.0`; full tool-surface snapshot coverage remains a future expansion beyond the representative BAT-007 scope. - Browser-level UI evidence criteria are captured in REQ-004 and remain local evidence rather than a required CI gate unless a future REQ changes that policy. - CI tests/lint required-check contract is defined for the six stable jobs listed above; actual branch protection enforcement remains an operator-managed repository setting. - BAT-008 adds read-only asset valuation MCP tools. Mutation MCP for manual asset CRUD or snapshot creation remains out of scope/future work. ## Change Log - 2026-06-26 22:24 KST: Created by `bootstrap-baseline` from repository observation. - 2026-06-27 10:17 KST: BAT-002/REQ-002 기준으로 backup/status/list/hash MCP contract와 restore 비노출 정책 추가. - 2026-06-29 KST: REQ-003/BAT-003 기준으로 account statement status update와 balance reconciliation report contract 추가. - 2026-06-29 KST: REQ-004/BAT-004 기준으로 Streamlit browser evidence 산출물과 UI evidence gate 규칙 추가. - 2026-06-29 KST: REQ-005/BAT-005 기준으로 account `liquidity_bucket`, `get_asset_liquidity_report`, dashboard operations hub contract 추가. - 2026-06-30 KST: REQ-007/REQ-008/BAT-007 기준으로 MCP response contract `1.0` governance, representative contract-regression scope, `ruff` lint, six-job required-check contract를 추가. - 2026-06-30 KST: REQ-009/REQ-010/BAT-008 기준으로 `list_manual_assets`, `get_asset_snapshot_report`, `get_net_worth_timeseries` read MCP contract와 dashboard valuation alignment 규칙을 추가.