Files
accounted/extensions/general/mcp-server
Jakob WennbergandClaude Fable 5 8e8b63a200 fix(bookkeeping): honor underlag VAT via vat_amount override in categorize flow (#717)
* fix(bookkeeping): honor underlag VAT via vat_amount override in categorize flow

The categorize flow always derived VAT as rate × gross/(1+rate) from the
transaction amount, with no way to use the underlag's actual moms. On e.g.
a restaurant receipt with dricks (no VAT on the tip), the agent could see
the document's correct VAT but the staged booking recomputed the wrong
rate-based amount on every attempt.

- buildMappingResultFromCategory: optional vatAmountOverride replaces the
  rate-derived VAT line ("Ingående/Utgående moms (enligt underlag)"; 0 =
  no VAT line). Rejects negatives, amounts above the 25%-extraction bound,
  and combination with reverse_charge / VAT-less treatments / private.
- gnubok_categorize_transaction: new vat_amount input, threaded into the
  staged preview and persisted in the operation params.
- commitCategorizeTransaction: reads params.vat_amount so the approved
  posting matches the staged preview exactly.
- PATCH /api/pending-operations/[id]: accepts vat_amount (null clears);
  preserves a staged override across category edits while the treatment
  still carries rate-based VAT, drops it when it no longer does.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* review: guard order + agent guidance on vat_amount (PR #717 bots)

- Check treatment compatibility before the 25%-extraction bound so an
  oversized override on reverse_charge reports the actual mistake (the
  treatment), not the amount. Document why the typeof re-check stays:
  commit-time params come from jsonb, so TS types don't hold at runtime.
- vat_amount property description now warns that foreign VAT is never
  deductible as ingående moms and that a 0-moms document should use
  vat_treatment="exempt" rather than vat_amount=0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(mcp): tools/list payload budget + reject vat_amount 0

core-only failed: the verbose vat_amount descriptions pushed the projected
tools/list payload to 36,051 tokens (ceiling 36,000; main is at 35,862).
Per the guard's own guidance, trim descriptions instead of bumping:
now 35,943.

Folds in the Swedish review's round-2 point while trimming: vat_amount 0
is now rejected with a pointer to vat_treatment "exempt". A 0-moms
document is an exempt supply — "exempt" produces the identical expense
booking and the correct income account (3004), so 0 had no use case and
only created a silent momsdeklaration misclassification path. Schema
declares exclusiveMinimum: 0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(bookkeeping): use roundOre for vat_amount math (antipattern ratchet)

Second core-only failure: the naive-ore-round ratchet caught the new
Math.round(x*100)/100 lines (662 > baseline 661). Switch the override
path to roundOre from lib/money — including the pre-existing computed-VAT
line this PR touched — and ratchet the baseline down (659, raw-route-auth
168 locked in from main-side fixes).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-06-12 11:42:21 +02:00
..
2026-05-04 11:12:29 +02:00

Accounted MCP server

JSON-RPC 2.0 server exposing the Accounted bookkeeping engine to MCP clients (Claude Desktop, Claude Code, etc.). Endpoint: /api/extensions/ext/mcp-server/mcp. OAuth and stdio bridge live alongside the API surface — see app/api/mcp-oauth/ and packages/gnubok-mcp/.

Tool authoring contract

Enforced by tests in __tests__/ — these are not style preferences, they're guard rails.

  1. additionalProperties: false on every inputSchema. Guarded by strict-schemas.test.ts. Forces clear rejections on hallucinated fields instead of silent ignores.
  2. Descriptions ≤ 280 chars. Guarded by output-schema.test.ts. No Args: / Returns: / Examples: prose — those belong in JSON Schema. Use agent-native hints ("Use to…", "Call X first", "HIGH risk").
  3. Staged-operation envelope for write tools — outputSchema: STAGED_OPERATION_SCHEMA (server.ts). Fields: staged, risk_level, actor, message, preview, period_status?, next?. The staged: true boolean is the explicit completion signal; agents must not infer completion from prose. Do NOT introduce a parallel { success, shouldContinue, output } envelope.
  4. period_status threading — any tool that ties to a fiscal-period-bound date (categorize, mark paid, create voucher, correct/reverse entry, approve supplier invoice) passes dateForPeriodCheck to stagePendingOperation. Response then includes period_status: { period_id, status: open|locked|closed, lock_date } so widgets and agents disable writes without round-trips.
  5. Scope mapping — every new tool needs an entry in lib/auth/api-keys.ts TOOL_SCOPE_MAP. Missing entries default to deny.
  6. Tests for new write tools — add staging-gate coverage to __tests__/voucher-tools.test.ts (or a sibling) plus executor coverage to lib/pending-operations/__tests__/voucher-executors.test.ts if the tool stages a new operation_type.

Determinism / cache stability

Tool definitions (name, description, inputSchema, outputSchema, annotations) are declared as static object literals at module load — no timestamps, no UUIDs, no Date/Math.random in the definition layer. This makes the tools/list JSON payload byte-stable across requests, which lets agent-side prompt caches stay warm. Do not introduce per-request non-determinism into the definitions block. Anything time-bound or random belongs inside execute().

For internal Anthropic API usage (today only extensions/general/invoice-inbox/lib/extract-invoice-fields.ts): annotate stable prefixes with cache_control: { type: 'ephemeral' } and log usage.cache_read_input_tokens for hit-ratio observability. The 1h TTL from the agent-native API plan (item 10) requires the direct Anthropic API; Accounted's Bedrock path defaults to a shorter TTL.

Payload-size watchdog

payload-size.bench.test.ts enforces a tools/list JSON payload ceiling (currently 36,000 tokens — bumped from 32,000 when top-level Tool.title landed on all tools for Claude Connectors Directory readiness). If the test fires, the right answer is rarely "raise the ceiling" — instead, trim descriptions or leverage gnubok_search_tools (already deployed; tool definitions can defer to it for discovery rather than enumerating in tools/list).

Where things live

  • server.ts — the tools array + JSON-RPC dispatcher
  • tool-result.ts — withNext(), toToolError() response helpers
  • resources/ — read-only Accounted:// URIs (active company, period, recent activity, capabilities, attention items, voucher gaps, chart of accounts, VAT treatments)
  • widgets/ — inline HTML widgets (receipt-matcher, vat-review)
  • prompts/ — slash-command-style prompts
  • skills/ — domain-knowledge skill bodies served via gnubok_load_skill
  • __tests__/ — strictness guards + per-tool coverage