fix(mcp): explain the Claude-side steps after "Anslut till Claude" and tick the checklist on a real connection (#2133) (#2147)
* fix(mcp): explain the Claude-side steps after "Anslut till Claude" and tick the checklist on a real connection (#2133) Lazy auth is by design: Claude lists the tools before any sign-in and the first company-scoped call answers 401, which opens the Accounted sign-in. Nothing told the user, so a "connected" status with an unanswered first question read as a broken connection (Axel, Discord). - Settings -> API & MCP: one sentence of expectation under the button, and the step-by-step guide link moved from under two disclosures to directly under the button. - Docs (connect-claude / anslut-claude): new "What happens after you click" section for Path A covering the connector dialog, the tools appearing before sign-in, the first-call login + consent screen, "ask again", and the "Required when the server asks" auth setting that only the manual path mentioned. - Hem checklist step "Anslut till Claude": deep link now carries client=claude-connector like the settings button (claudeConnectorLink), the footnote carries the same expectation line plus the guide link, and the done-signal is an unrevoked api_keys row minted by the MCP OAuth token route (OAUTH_MCP_KEY_NAME) instead of the in-app AI-profile flag, which never meant "connected to Claude". - Tests: claudeStepDone with/without a key row, deep-link snapshot. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W7iJQwKiRTDWSMnRm4WM4L * fix(mcp): correct consent-page claims, stop the completion PATCH loop, count OAuth keys past RLS (#2133) Three skeptic refutations on PR #2147, fixed in one pass: - Docs (EN + SV): the consent page shows the company active in the app and pre-selects every scope for Claude's connector (founder decision 2026-08-26); it has no company picker and nothing to tick. Steps 3-4 of the new section, the "Read-only by default" paragraph above it, the sandbox note and the 10-minute test now describe Endast läs under Behörigheter instead. - Checklist completion: users with initial_setup_path NULL (skipped the books question, then imported) hit the route's "Välj först hur du vill komma igång" 400 and, with saving as an effect dependency, retried it forever with a toast. completionPatchBody() records path=migration when none was chosen, and a rejected PATCH is not retried within the session. - hasMcpKey: api_keys' SELECT policy is company-scoped, so the user client could not see companyless (NULL company_id) or archived-company keys and the step stayed open for the user who had just connected. The head count now runs through the service client with an explicit user_id filter. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W7iJQwKiRTDWSMnRm4WM4L * fix(mcp): surface a failed OAuth-key count and reserve the marker name (#2133) CodeRabbit round on PR #2147: - app/(dashboard)/page.tsx: a failed api_keys count answered count null, which claudeStepDone read as "never connected". Throw to the error boundary like the settings fetch does instead of guessing. - app/api/settings/api-keys: reject a hand-minted key named MCP-klient (OAuth) (400 VALIDATION_ERROR): that name is the marker the Hem checklist reads as "connected to Claude", so a manual key with it would tick the step without any connection. Test added. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01W7iJQwKiRTDWSMnRm4WM4L --------- Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Fable 5.1
parent
8b09b06e14
commit
4f33184a9a
@@ -1475,6 +1475,7 @@ One line per decision: `[YYYY-MM-DD] <decision>: <why>`. Appended by agents and
|
||||
[2026-09-01] mcp.tool_called gets errorCause = errorCauseTag(err) on the two execution catch paths only (#2051): SQLSTATE or coded-error code, else the error class name, capped at 64 chars; a plain Error deliberately tags null because the class name 'Error' is noise, and pre-execution denials pass nothing since their errorCode already IS the vocabulary. Raw driver messages stay out of event_log on purpose: a constraint-violation message can quote row values.
|
||||
[2026-09-01] counterparty_aliases joins the categorization_templates audit-trigger strip list (20260901200000) instead of staying logged: prod falsified the original exclusion list within 30 minutes of 20260901103000 going live (15 of the first 16 UPDATE audit rows were alias+learning noise, ~800/day projected vs ~50/day of real rule changes), because the learning path merges aliases in the same write that bumps occurrence_count. Explicit trade-off: a human editing ONLY aliases is no longer logged; accepted since alias growth is overwhelmingly automatic and any change also touching accounts/VAT/pattern/active still logs (first real one, 19:02:17Z same day, captured correctly). Pre-fix noise rows stay in audit_log (append-only) and the read model stops labelling the column so they render as no-ops.
|
||||
[2026-09-01] MCP catalog budget attacked at the duplicated staged envelope rather than by demoting more reads: measuring the payload by segment showed outputSchema is 38 % of the whole catalog (23 290 tokens) and STAGED_OPERATION_SCHEMA alone 14 736 of it, the same envelope transmitted 58 times, while descriptions (what the three previous rounds trimmed) are only 10 %. period_status now carries its shape in one sentence instead of declared JSON Schema, matching actor/approve/preview which were always bare objects; 2 552 tokens reclaimed with no tool demoted and no field removed. Every edit is in the LOOSER direction because the server emits structuredContent for every tool and the documented failure mode is a declaration too tight making a strict client reject a successful call. next kept additionalProperties: false: staging.test.ts pins it closed and a guard whose reason is not in front of you is not one to loosen for 420 tokens. Ceiling ratcheted to 60 000 rather than the usual ~300 margin, leaving ~1 070 deliberate working margin: server.ts took 70 commits in 14 days and the previous 116-token margin is what starts the ratchet-block-bump-demote cycle visible in the bench log.
|
||||
[2026-09-01] Checklist "Anslut till Claude" done-signal = unrevoked api_keys row named MCP-klient (OAuth) for the USER, not per company: the OAuth token route is the only writer of that name and the key company_id is whatever was active at sign-in (null for companyless signups), so a company filter would miss real connections; the AI-profile flag it replaced never meant "connected to Claude" (#2133). Counted through the service client with an explicit user_id filter, not the user client: api_keys' SELECT policy is company-scoped (20260330130000), which hides companyless and archived-company keys and left the step open for exactly the user who had just connected (skeptic refutation on PR #2147). The manual create route reserves the name (400) rather than adding a source column: a migration for a cosmetic tick is not worth it, and the name is already the only marker every reader of that row uses. The consent-page default (all scopes pre-selected, founder decision 2026-08-26) is described, not changed; the compliance swarm's GDPR Art.25(2) finding on this PR targets that decision and is Emil's call, not this docs fix.
|
||||
[2026-09-01] PR #2130 CodeRabbit P1 (org-number twin inherits a grant): the ombud path binds SKV system-credential access to an org number, and org numbers are public and tenant-editable, so while more than one live (non-archived) company claims the same 12-digit org number NO company may verify, mint a deep link, or be granted by the nightly sync on it (409 ORG_NUMBER_CONTESTED; cron counts them as contested and changes nothing). This does not re-add the company-creation org-number guard (org-number reuse stays allowed); it only fences the one feature where the org number is the authority boundary. Also: the cron honours summarizeGrants.recognized (unknown role codes = pinning problem, never a denial), mirroring probeViaOmbudsregister.
|
||||
|
||||
[2026-09-01] ENABLE_BANKING_SANDBOX removed from the enable-banking manifest and the index.ts header (#2131): the variable was declared as optional but never read anywhere; sandbox vs production is decided by ENABLE_BANKING_API_URL (api.tilisy.com vs api.enablebanking.com, api-client.ts derives isSandbox from the host). A dead variable declared in the manifest is what the self-hosting docs would otherwise have copied. The manifest now lists the two optional variables the code actually reads (API_URL, PSU_TYPE); the _PRODUCTION aliases stay undeclared on purpose, they are a hosted Vercel convention, not an operator contract.
|
||||
|
||||
Reference in New Issue
Block a user