Files
accounted/lib/docs/content/connect-claude.ts
T
Jakob Wennberg 0521c385d2 feat(transactions): underlag status badges + attach dialog; auto-expire stale pending ops (#712)
* feat(transactions): per-row underlag status + attach-document dialog

- New "Matcha mot underlag" dialog on /transactions (inbox pick or fresh
  upload), the tx→doc mirror of the Documents view's matcher
- Per-row Underlag/Underlag saknas badges on booked history rows, driven
  by computeJeUnderlagStatus — same posted-only, exemption-aware scope as
  the worklist count so badge and count never disagree
- attach-document route + commit dispatcher now propagate the doc onto
  the verifikation when the tx is already booked (BFL 5 kap 6 §), with a
  409 guard for docs consumed by a different verifikation, idempotent
  re-attach (no same-value rewrite under period lock), and an honest 409
  when the period-lock trigger blocks the propagation
- Booking-dialog doc links also pin the doc to the transaction row
  (first linked doc wins) via the link route's new transaction_id param

messages/{sv,en}.json also carries the strings for the pending-ops
expiry UI that lands in the next commit.

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

* feat(pending-operations): auto-expire stale staged operations after 30 days

- New daily cron (02:30 UTC, vercel.json + both docker crontabs) flips
  >30-day-old pending ops to rejected with the dispatcher's
  { auto_rejected: true, reason: 'expired' } result_data shape — rows are
  never deleted, the table is the audit trail
- /pending renders an "Utgick automatiskt" badge + detail line for these,
  orders terminal tabs by resolved_at so a fresh expiry sweep isn't
  buried, and adds a first-time-reviewer explainer
- Origin labels spell out where a proposal came from (AI chat, MCP key,
  API, cron) instead of the raw actor_label
- agent_chat actor type added to PendingOperationActorType/AuditLogEntry
  (DB CHECK already widened in 20260519090000) and to the agent filter
- ApprovalCard notes that ignoring a proposal is safe

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

* docs(mcp): surface the client telemetry marker in connect instructions

Tag the connector URLs shown in ApiKeysPanel, the connect-claude doc and
the gnubok-mcp README with ?client=<surface> (claude-connector /
claude-code) and GNUBOK_CLIENT=claude-desktop for the npm bridge.
Telemetry-only — the server already reads the param/header; this just
lets us measure which Claude surface connected.

The claude mcp add copy blocks quote the URL: an unquoted ? in the query
string trips zsh globbing ("no matches found").

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

* review: fix stale-closure badge flip + zod-validate link route body (PR #712)

- handleDocumentAttached read journal_entry_id off the render-time
  transactions snapshot; if the list changed while the attach dialog was
  open the optimistic badge flip was silently skipped. Read it off the
  dialog's own subject (attachDocTx) instead.
- POST /api/documents/[id]/link now validates the body against the new
  LinkDocumentSchema (uuid-strict, all four fields) instead of a bare
  presence check on journal_entry_id — same canonical VALIDATION_ERROR
  envelope. Test fixtures switched to real UUIDs accordingly.

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

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-06-11 11:13:51 +02:00

81 lines
5.8 KiB
TypeScript

export const CONNECT_CLAUDE_MD = `# Connect with Claude
> Talk to your bookkeeping. Connect Accounted to Claude (claude.ai, Claude Desktop, or Claude Code) and ask questions, categorise transactions, and prepare a momsdeklaration in plain language — every write still stages for your approval first.
Accounted ships an [MCP](https://modelcontextprotocol.io) server that exposes the full bookkeeping engine — 90+ tools — to any MCP client. The endpoint is:
\`\`\`
https://app.gnubok.se/api/extensions/ext/mcp-server/mcp
\`\`\`
There are two ways to connect, depending on your client. Both reach the same tools and the same approval model: read tools answer immediately, write tools (categorise, mark paid, create voucher, year-end) **stage a pending operation** that you confirm in chat or in the **/pending** web UI before anything is booked.
## Path A — claude.ai / Claude Desktop custom connector (OAuth 2.1)
Best for most users. No API key to manage — you authorise Accounted the same way you'd authorise any OAuth app.
1. In **claude.ai** (Settings → Connectors) or **Claude Desktop** (Settings → Connectors → Add custom connector), choose **Add custom connector**.
2. Paste the connector URL:
\`\`\`
https://app.gnubok.se/api/extensions/ext/mcp-server/mcp?client=claude-connector
\`\`\`
_(The "?client=claude-connector" suffix is telemetry-only — it lets Accounted see you connected via claude.ai/Desktop. It changes nothing about behaviour or scopes; drop the query string if you prefer.)_
3. Claude opens the Accounted OAuth 2.1 consent screen. Sign in and pick the company you want Claude to act on.
4. On the consent screen you grant **read-only scopes by default** (list invoices, read reports, compute VAT). Write scopes (create invoice, categorise, book vouchers, run year-end) are **listed separately and must be ticked explicitly** — leave them unchecked for a read-only review session.
5. Approve. Claude now lists the Accounted tools and you can start asking questions.
Because the consent is per-company and scoped, you can connect a read-only key for a reviewer and a separate write-enabled connection for day-to-day bookkeeping.
## Path B — \`npx gnubok-mcp\` with an API key (stdio bridge)
Best for Claude Desktop on a machine where you'd rather use a long-lived API key than the OAuth flow, or for scripting.
1. Mint an API key in the Accounted dashboard at **/settings/api**. Use a \`gnubok_sk_test_*\` key against the sandbox while you evaluate; switch to \`gnubok_sk_live_*\` for real data.
2. Add the stdio bridge to your \`claude_desktop_config.json\`:
\`\`\`json
{
"mcpServers": {
"gnubok": {
"command": "npx",
"args": ["gnubok-mcp"],
"env": {
"GNUBOK_API_KEY": "gnubok_sk_test_...",
"GNUBOK_CLIENT": "claude-desktop"
}
}
}
}
\`\`\`
3. Restart Claude Desktop. The bridge proxies stdio JSON-RPC to the hosted MCP endpoint over HTTPS; the key carries the scopes you granted it at mint time.
The key's scopes gate exactly which tools are callable — a key without write scopes can read reports and ledgers but cannot stage a booking.
## Try these prompts
All three run against the deterministic sandbox seed (use a \`gnubok_sk_test_*\` key or pick the sandbox company on the OAuth consent screen). They exercise the read path end-to-end without booking anything.
1. **"Show my uncategorized bank transactions and suggest categories."**
Claude calls \`gnubok_list_uncategorized_transactions\` then \`gnubok_suggest_categories\` and walks you through the proposals. Approving one stages a \`gnubok_categorize_transaction\` pending operation — nothing is booked until you confirm.
2. **"Which invoices are overdue?"**
Claude calls \`gnubok_get_ar_ledger\` (kundreskontra) and lists outstanding customer invoices with aging.
3. **"Compute my VAT report for this quarter and tell me if I can close it."**
Claude calls \`gnubok_get_vat_report\` for the momsdeklaration rutor, then \`gnubok_vat_close_check\` to scan for blockers (uncategorised rows, unapproved supplier invoices, missing receipts on expenses ≥ 4 000 kr — the tool's high-value heuristic; BFL requires underlag for every affärshändelse regardless of amount) and reports \`ready_to_close\`.
## 10-minute reviewer test
A quick end-to-end pass to confirm the connection works before you trust it with real data. Run the steps in order; each lists what you do and what you should see.
1. **Connect.** Use Path A (read-only scopes only) or Path B with a \`gnubok_sk_test_*\` key. → Claude lists the Accounted tools (titles like *List Uncategorized Transactions*, *VAT Declaration (Momsdeklaration)*).
2. **Confirm the company.** Ask *"Which company am I connected to?"* → Claude names the sandbox company (e.g. **Sandlådan Konsult**).
3. **Run prompt 1** (*uncategorized + suggest categories*). → A list of uncategorised rows plus category suggestions; no booking happens.
4. **Run prompt 2** (*overdue invoices*). → At least one overdue customer invoice with aging.
5. **Run prompt 3** (*VAT report + can I close*). → Momsdeklaration rutor returned; \`vat_close_check\` reports a **non-empty blocker list** (uncategorised transactions, an unapproved leverantörsfaktura, and a high-value business expense without a receipt).
6. **Stage a write.** Ask Claude to categorise one transaction. → Claude stages a pending operation and asks you to confirm — the booking does **not** post until you approve in chat or at **/pending**.
If every step matches, the connector is wired correctly and the approval model is enforced.
## Support
Stuck connecting, or seeing an unexpected blocker? Use the in-app support form at **/help** — it routes straight to the product team with your company context attached. Include the client (claude.ai / Desktop / Code), the path you used (A or B), and the tool name from any error message.
`