@kaminari-ad/mcp 0.13.0 → 0.14.1

package/CHANGELOG.md

@@ -7,6 +7,114 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.14.1] - 2026-08-24
+
+### Changed
+
+- **Three tag tools described archived tags as still reachable
+  (KAMIAD-120).** The API now retires an archived tag everywhere: it is
+  no longer assigned to scans, no longer raises alerts, and an archived
+  platform tag is gone from the catalogue entirely. `list_tags` told the
+  agent the opposite — to pass `include_archived` "when resolving a slug
+  seen on an older scan", which now returns nothing for exactly those
+  slugs. `get_tag_definition` now documents the 404, and
+  `list_scan_tags` that a scan can report fewer tags than when it ran.
+  Descriptions only; no schema or shape change, so
+  `src/shared/api/{openapi,zod-schemas}.ts` are unaffected — the two
+  stale JSDoc lines about `include_archived` are regenerated from the
+  live spec once the API side is deployed.
+
+## [0.14.0] - 2026-08-23
+
+Resyncs the tool surface with `/api/v1` (KAMIAD-158). The generated
+types were already current; the hand-written gateway and tool schemas
+were not, so this release is mostly about the drift between them — plus
+two gates so the next one fails CI instead of reaching a user.
+
+### Fixed
+
+- **Artifact downloads were unbounded.** `binaryGet` called
+  `arrayBuffer()` with no ceiling, so one oversized MediaFile decided
+  how much memory the hosted server used — the API caps none of these
+  endpoints. It now refuses on an oversized `content-length` and
+  counts bytes while reading the stream, cancelling past the limit
+  (256 KiB text, 8 MiB binary). The cap covers the screenshot and
+  invoice-PDF tools too, which had the same exposure.
+- **`attach_policy_set_campaigns` described a side effect it does not
+  have.** A campaign belongs to exactly one policy set, so attaching
+  one that is bound elsewhere MOVES it; the tool said existing
+  bindings survive. Now flagged `destructiveHint: true` and the
+  description says so.
+- **`bulk_update_alert_status` forwarded `filter_*` alongside `ids`**,
+  which the API rejects. Refused locally with the offending field
+  names.
+- **`list_campaign_groups` advertised `last_run_at`** while the
+  response projection stripped it, so the new last-run filters could
+  not be interpreted. Projected now.
+- **`set_campaign_alert_overrides` accepted values the API rejects.**
+  `mode` was `inherit | include | exclude`; the API has only ever
+  accepted `inherit | override | silence`. Every value the schema
+  allowed was refused upstream and every value the API wanted was
+  refused locally, so an agent could not change campaign alert routing
+  at all. `mode` now matches the API, `destination_ids` is refused
+  locally unless `mode` is `override` (mirroring the API's
+  `notifications.invalid_override_combination`), and the setter's
+  description explains that "route everywhere except these" does not
+  exist — invert the list under `override` instead.
+- **`list_custom_taxonomies` claimed soft-deleted taxonomies were
+  listed.** They are excluded unless the new `include_inactive` is
+  `true`, so an agent looking for one to restore concluded it was gone.
+- `list_scans` now documents `rechecking` as a filterable status.
+
+### Added
+
+- **`bulk_update_alert_status`** — `POST /api/v1/alerts/bulk-status`.
+  Select with `ids` (max 1000) or `all_matching: true` plus `filter_*`
+  fields; returns `updated` / `skipped`.
+- **`list_policy_set_campaigns`**, **`attach_policy_set_campaigns`**,
+  **`detach_policy_set_campaigns`** — policy-set membership, which was
+  previously not editable through MCP at all: `update_policy_set` sends
+  only name, description and rules. Attach warns that a campaign bound
+  to another set is moved rather than rejected, since a campaign can
+  belong to only one set.
+- **`get_scan_creative_html`**, **`get_scan_vast_xml`** — creative
+  markup and the resolved VAST document, returned as text the model
+  reads directly. **`get_scan_creative_video`** — the MP4 as a resource
+  block. All three refuse oversized payloads (256 KiB text, 8 MiB
+  binary) rather than inlining them.
+- Filters the API had grown and the tools never picked up:
+  `list_alerts` and `get_alert_stats` take `policy_set_id`, `tag`,
+  `country_code`, `date_from`, `date_to`, `timezone` (`get_alert_stats`
+  previously took no arguments at all, so its counts could not be
+  reconciled with a filtered list); `list_scans` takes `tag_match` and
+  `parent_scan_id`; `list_campaigns` and `list_campaign_groups` take
+  `created_from`/`created_to`/`last_run_from`/`last_run_to`/`timezone`,
+  and `list_campaign_groups` also takes `q`; `list_tags` takes
+  `include_archived`.
+- **`check:api-coverage`** — asserts every `/api/v1` path + method in
+  the generated `openapi.ts` is reachable from the gateway, and that
+  each of their query parameters is representable by the gateway
+  method. The parameter half is the point: a path-only gate would have
+  passed on the state this release fixes.
+- **`check:tool-enum-drift`** — every `z.enum` in a tool must have the
+  same value set as a generated enum, or carry a justified exemption.
+  Reintroducing the `mode` bug now fails CI. Equality rather than
+  subset: with 68 values across the generated enums, "subset of any"
+  accepted almost anything (`z.enum(["open"])` passed as a subset of
+  `AlertStatus`). A `z.enum` whose argument is not an inline array of
+  literals is reported rather than skipped, since an unreadable enum
+  is how a wrong value set would slip past.
+
+### Changed
+
+- `npm run gen:api-types` formats its output with Prettier. The raw
+  generator emits 4-space TypeScript while `format:check` covers
+  `src/**`, so a no-op regen used to produce a ~17k-line indentation
+  diff and a failing gate.
+- `.prettierignore` no longer lists `src/infrastructure/api/openapi.ts`
+  — the generated files moved to `src/shared/api/` long ago, so the
+  entry was dead and pointed at the wrong policy.
+
 ## [0.13.0] - 2026-08-20
 
 ### Added
@@ -1037,7 +1145,9 @@ Initial public release. The first version that ships to npm under
   need them.
 - Invoice PDF fetcher — same reason.
 
-[Unreleased]: https://github.com/kaminari-ad/mcp/compare/v0.13.0...HEAD
+[Unreleased]: https://github.com/kaminari-ad/mcp/compare/v0.14.1...HEAD
+[0.14.1]: https://github.com/kaminari-ad/mcp/compare/v0.14.0...v0.14.1
+[0.14.0]: https://github.com/kaminari-ad/mcp/compare/v0.13.0...v0.14.0
 [0.13.0]: https://github.com/kaminari-ad/mcp/compare/v0.12.0...v0.13.0
 [0.12.0]: https://github.com/kaminari-ad/mcp/compare/v0.11.0...v0.12.0
 [0.11.0]: https://github.com/kaminari-ad/mcp/compare/v0.10.0...v0.11.0

package/README.md

@@ -63,7 +63,7 @@ Add to your MCP client config (Cursor: `~/.cursor/mcp.json`; Claude Desktop: `~/
 }
 ```
 
-Restart your client. You should see `kaminari-ad` in the MCP servers list with 98 tools exposed.
+Restart your client. You should see `kaminari-ad` in the MCP servers list with 105 tools exposed.
 
 ### 2b. Hosted HTTP transport (no install)
 
@@ -113,34 +113,34 @@ which decides which credential type minted it.
 
 ## Tools
 
-98 tools mirroring most of the public `/api/v1` surface of Kaminari Ad. Every tool carries MCP behaviour annotations (`title`, `readOnlyHint`, `destructiveHint`, `idempotentHint`, `openWorldHint`) so MCP clients can warn before destructive actions. The complete list, by domain:
+105 tools covering the public `/api/v1` surface of Kaminari Ad. Every tool carries MCP behaviour annotations (`title`, `readOnlyHint`, `destructiveHint`, `idempotentHint`, `openWorldHint`) so MCP clients can warn before destructive actions. The complete list, by domain:
 
 - **Account** (14) — `get_account`, `update_org`, `list_org_users`, `invite_user`, `update_user_role`, `remove_user`, `transfer_ownership`, `list_org_roles`, `create_custom_role`, `list_account_labels`, `update_account_labels`, `list_api_keys`, `create_api_key`, `revoke_api_key`
-- **Scans** (10) — `list_scans`, `get_scan`, `list_scan_children`, `create_scan`, `create_bulk_scans`, `recheck_scans`, `cancel_scan`, `get_scan_screenshot`, `get_scan_creative_screenshot`, `get_scan_landing_screenshot`
+- **Scans** (13) — `list_scans`, `get_scan`, `list_scan_children`, `create_scan`, `create_bulk_scans`, `recheck_scans`, `cancel_scan`, `get_scan_screenshot`, `get_scan_creative_screenshot`, `get_scan_landing_screenshot`, `get_scan_creative_html`, `get_scan_creative_video`, `get_scan_vast_xml`
 - **Campaigns** (10) — `list_campaigns`, `list_campaigns_picker`, `get_campaign`, `create_campaign`, `update_campaign`, `archive_campaign`, `unarchive_campaign`, `cancel_campaign`, `run_campaign`, `list_campaign_runs`
 - **Campaign groups** (10) — list/get/create/update/run/cancel/archive/unarchive + `pause_campaign_group_schedule`, `resume_campaign_group_schedule`
 - **Runs** (3) — `get_run`, `list_run_scans`, `cancel_run` (use `list_campaign_runs` to enumerate runs of a campaign — the API has no standalone `/runs` index)
 - **Tags** (5) — `list_tags`, `get_tag_definition`, `update_tag_definition`, `delete_tag_definition`, `list_scan_tags`
 - **Custom rules** (6) — `list_custom_rules`, `get_custom_rule`, `create_custom_rule`, `update_custom_rule`, `delete_custom_rule`, `test_custom_rule`
 - **Custom taxonomies** (7) — `list_custom_taxonomies`, `get_custom_taxonomy`, `create_custom_taxonomy`, `update_custom_taxonomy`, `delete_custom_taxonomy`, `restore_custom_taxonomy`, `parse_custom_taxonomy_text`
-- **Policy sets** (6) — `list_policy_sets`, `get_policy_set`, `create_policy_set`, `update_policy_set`, `delete_policy_set`, `request_policy_set_approval`
-- **Alerts** (3) — `list_alerts`, `update_alert_status`, `get_alert_stats`
+- **Policy sets** (9) — `list_policy_sets`, `get_policy_set`, `create_policy_set`, `update_policy_set`, `delete_policy_set`, `request_policy_set_approval`, `list_policy_set_campaigns`, `attach_policy_set_campaigns`, `detach_policy_set_campaigns`
+- **Alerts** (4) — `list_alerts`, `update_alert_status`, `bulk_update_alert_status`, `get_alert_stats`
 - **Webhooks** (11) — `list_webhooks`, `get_webhook`, `create_webhook`, `update_webhook`, `delete_webhook`, `list_webhook_event_types`, `list_webhook_deliveries`, `test_webhook`, `rotate_webhook_secret`, `replay_webhook_delivery`, `bulk_replay_webhook`
 - **Billing** (4) — `get_billing_summary`, `list_usage`, `get_usage_summary`, `list_balance_history`
 - **Invoicing** (2) — `list_invoices`, `get_invoice_pdf`
 - **Alert notifications** (5) — `list_alert_destinations`, `delete_alert_destination`, `set_alert_destination_version`, `get_campaign_alert_overrides`, `set_campaign_alert_overrides`
 - **Reference data** (2) — `list_geos`, `list_emulators`
 
-Screenshots (`get_scan_screenshot`, `get_scan_creative_screenshot`, `get_scan_landing_screenshot`) come back as inline MCP `image` blocks and `get_invoice_pdf` as an inline resource block — no second fetch, no presigned URL.
+Screenshots (`get_scan_screenshot`, `get_scan_creative_screenshot`, `get_scan_landing_screenshot`) come back as inline MCP `image` blocks; `get_invoice_pdf` and `get_scan_creative_video` as inline resource blocks — no second fetch, no presigned URL. The two text artifacts (`get_scan_creative_html`, `get_scan_vast_xml`) come back as strings the model can read directly. Every artifact download is size-capped in the gateway — 256 KiB for the text artifacts, 8 MiB for the binary ones — and refused while reading rather than buffered and then rejected.
 
-Not exposed (intentionally): the raw creative artifacts (`creative-html`, `creative-video`, `vast-xml`) and the public marketing forms (`/contact`, `/demo-inquiries`). Open an issue if you need one of those.
+Not exposed (intentionally): the public marketing forms, which are anonymous intake for kaminari.ad itself rather than an agent capability.
 
 ## Example agent prompts
 
 These three prompts each exercise a different cross-section of tools and demonstrate the typical agent workflow:
 
 1. **"Scan https://news.example.com/article-promo across US, UK, DE on mobile profiles, flag anything that redirects to a paywall."** Touches `list_emulators` → `create_bulk_scans` → wait → `list_scans` (status=completed) → `get_scan` → `list_scan_tags`.
-2. **"Create a campaign that re-checks the homepage of brand-x.com every hour from JP and US; alert me on Slack if it ever shows a malware tag."** Touches `list_emulators` → `list_policy_sets` (find one with `malware`) → `create_campaign` (schedule_enabled=true) → `list_alert_destinations` → `set_campaign_alert_overrides`.
+2. **"Create a campaign that re-checks the homepage of brand-x.com every hour from JP and US; alert me on Slack if it ever shows a malware tag."** Touches `list_emulators` → `list_policy_sets` (find one with `malware`) → `create_campaign` (schedule_enabled=true) → `attach_policy_set_campaigns` → `list_alert_destinations` → `set_campaign_alert_overrides` (`mode: "override"` with the Slack destination).
 3. **"What did I spend on ad verification last month, and which campaigns drove the cost?"** Touches `get_usage_summary` → `list_usage` (with date_from/date_to) → group by `scan_id` → `get_scan` → `get_campaign` for attribution.
 
 Full machine-readable tool listing is exposed by the server itself — connect with any MCP client and call `tools/list`.

package/dist/bin.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { NAME, VERSION, err, ok } from './chunk-VFDW3UL2.js';
+import { NAME, VERSION, err, ok } from './chunk-YS7IDW24.js';
 import process from 'process';
 import { z } from 'zod';
 
@@ -160,10 +160,10 @@ async function main() {
   }
   const config = configResult.value;
   if (config.transport === "stdio") {
-    const { bootstrapStdio } = await import('./stdio-bootstrap-VM6OOG2O.js');
+    const { bootstrapStdio } = await import('./stdio-bootstrap-II6IVJWD.js');
     return bootstrapStdio(config);
   }
-  const { bootstrapHttp } = await import('./http-bootstrap-ROIMZVK2.js');
+  const { bootstrapHttp } = await import('./http-bootstrap-XAA75RA3.js');
   return bootstrapHttp(config);
 }
 main().then(