@path58/p58-n8n 0.2.13 → 0.2.16

package/AGENT_INSTALL.md

@@ -8,7 +8,7 @@
 
 ## What is p58-n8n?
 
-p58-n8n is an MCP server that gives LLMs the ability to validate, build, deploy, and manage n8n workflows. It provides 12 tools in the default configuration (35 total registered, 11 offline-capable), organized in tiers:
+p58-n8n is an MCP server that gives LLMs the ability to validate, build, deploy, and manage n8n workflows. It provides 12 tools in the default configuration (34 total registered, 11 offline-capable), organized in tiers:
 
 - **Public/basic (offline-capable):** planning, node and operation discovery, credential schema discovery, and public-safe validation work immediately from bundled catalog assets
 - **Public/basic + n8n:** build, deploy, test, workflow CRUD, and credential CRUD work against the user's own n8n instance
@@ -316,4 +316,4 @@ Once installed, the user can ask their AI assistant to:
 
 ---
 
-**Package:** `@path58/p58-n8n` | **npm:** https://www.npmjs.com/package/@path58/p58-n8n | **Version:** 0.2.13+
+**Package:** `@path58/p58-n8n` | **npm:** https://www.npmjs.com/package/@path58/p58-n8n | **Version:** 0.2.15+

package/CHANGELOG.md

@@ -7,6 +7,67 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+
+- **`activate_premium` MCP tool** (RAG-4.63.7) — LLM-guided premium catalog activation. Accepts a P58 access token, validates it against VPS API (`POST /v1/auth/validate`), and persists to `~/.config/p58-n8n/config.json` with atomic write (write→rename). `setup_check` updated with activation guidance when no token is present.
+- **ProxyCatalogAdapter** (RAG-4.63.6) — Third routing path in `getCatalogAdapter()`: `P58_ACCESS_TOKEN` + no `P58_DATABASE_URL` → REST API proxy to VPS. Implements all 22 `CatalogAdapter` interface methods via HTTP fat node blobs + local `Map` cache. Bulk warm-up via `POST /v1/catalog/bulk` reduces API calls to ~5-10 per build session.
+- **VPS Catalog REST API** (RAG-4.63.4) — Express server at `p58-n8n.path58.com/v1/catalog/*` serving premium catalog data from Redis/Supabase. 5+1 endpoints, gzip compression, `connectionManager` integration.
+- **Token auth & rate limiting** (RAG-4.63.5) — Bearer token auth with in-memory SHA-256 hash set (5-min refresh from Supabase `premium_tokens`). Application-level rate limiting: 100 req/min/token.
+- **Telemetry & graceful degradation** (RAG-4.63.8) — Async request logging to Supabase `catalog_api_logs`. Telegram health alerts via n8n webhook. Client-side degradation wrapper: 3 consecutive failures → fallback to `PublicCatalogAdapter` with user notification, 30s recovery poll.
+- **Config persistence** (RAG-4.63.7) — Shared `p58-config.ts` with `readTokenFromConfig()`. Token resolution precedence: `P58_ACCESS_TOKEN` env → `.env` → `~/.config/p58-n8n/config.json`.
+
+## [0.2.15] - 2026-03-29
+
+### Fixed
+
+- **`SERVER_VERSION` was stale in v0.2.14** — `config.ts` still reported `0.2.13`. Now correctly reports `0.2.15`.
+- **README badge URL** — Shield.io URL-encoded portion still showed `9%2C840` while display text said 9,860. Both now consistent at 9,860.
+- **AGENT_INSTALL.md version** — Updated from `0.2.13+` to `0.2.15+`.
+
+### Changed
+
+- **CHANGELOG backfill** — Added missing entries for v0.2.12 and v0.2.14.
+
+## [0.2.14] - 2026-03-29
+
+### Added
+
+- **Credential health stats** — `setup_check` now returns `credential_health: { valid, invalid, untested }` summary from the credential cache.
+- **Product-language setup diagnostics** — `setup_check` fields renamed from internal jargon (`db_connected`, `catalog_ready`, `n8n_connected`) to user-facing terms (`enhanced_intelligence_available`, `intelligence_status`, `workflow_engine_connected`). All issue messages rewritten in product language.
+- **`probeWorkflowEngine()`** — `setup_check` now probes the actual n8n API (`/settings` endpoint) instead of checking whether the credential cache loaded.
+- **Clone safety guards** — Both `executeCloneTest()` (build-workflow) and `testViaClone()` (test-workflow) now check `hasRespondToWebhookNode()` before cloning; R2W workflows skip clone entirely.
+- **`retryWebhookTest()`** — New fallback in `runTestStep()`: when webhook-test POST gets 404, retries via production `/webhook/{path}` endpoint instead of UI-only `/webhook-test/{path}`.
+- **`test_passed` / `test_error` top-level fields** — `BuildWorkflowResponse` now includes explicit pass/fail signaling so LLMs never mistake a deployed-but-failing workflow for success.
+- **`isError: true` on test failure** — `buildTestSuccessResponse()` now checks `result.status === 'error'` and sets MCP `isError` flag with `⚠️ WORKFLOW DEPLOYED BUT TEST FAILED:` prefix.
+- **Fused parser `defaultParameters`** — All 13 service mappings in `descriptionToSpec.ts` now include required node parameters (chatId, text, channel, etc.) so deployed workflows actually function.
+- **Trigger-action keyword deduplication** — Fused parser no longer matches the same keyword as both trigger and action node.
+- **Clone poll error handling** — Clone execution polling no longer infinite-loops on errored executions.
+- **16 new unit tests** — `rag-4.58.24-ux-hardening.test.ts` covering all 6 deliverables.
+
+### Fixed
+
+- **Respond-to-Webhook P0 clone bug** — `runTestStep()` fell back to `executeCloneTest()` on webhook 404, but cloning R2W workflows causes n8n 500 errors. Now guards with `hasRespondToWebhookNode()` check and retries via direct POST.
+- **Test failure signaling** — `buildTestSuccessResponse()` always wrapped results with `success: true` even when tests failed. LLMs saw "success" and stopped investigating. Now correctly propagates failure.
+- **`n8n_connected` false positives** — Was checking `cacheStatus.loaded` (credential cache state), not actual n8n API reachability. Replaced with live API probe.
+
+## [0.2.12] - 2026-03-20
+
+### Added
+
+- **Config C fused build** — Single-turn plan+build via `build_workflow` with `mode: "fused"`. User describes a workflow in plain English; the server plans architecture, selects nodes, wires credentials, validates, and deploys — all in one tool call.
+- **Minimal build mode** — `build_workflow` accepts `mode: "minimal"` for lightweight workflows without the full planning pipeline.
+- **Blind validation** — 5 real-world validation scenarios (VT001–VT005) tested end-to-end against live n8n.
+- **P0 bug fix sprint** — 6 critical bugs from validation testing fixed (RAG-4.58.5).
+- **Webhook testing** — Enriched `test_workflow` with webhook discovery and automatic POST testing.
+- **AI Agent build quality** — Pattern detection for LangChain agent workflows with auto-fix loop.
+- **Execution error reporting** — Structured error extraction from n8n execution results.
+- **Code node auto-inject** — JavaScript and Python code nodes get correct boilerplate.
+
+### Fixed
+
+- **Webhook path extraction** — Multiple fixes for webhook URL construction in build and test paths.
+- **LangChain connection wiring** — AI sub-nodes now wire to correct diamond ports with proper connection types.
+
 ## [0.2.13] - 2026-03-28
 
 ### Added
@@ -209,7 +270,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - npm package published as `@path58/p58-n8n`
 - ESM module support with shebang for direct `npx` execution
 
+[0.2.15]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.15
+[0.2.14]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.14
 [0.2.13]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.13
+[0.2.12]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.12
 [0.2.7]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.7
 [0.2.6]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.6
 [0.2.5]: https://github.com/tsvika58/p58-n8n/releases/tag/v0.2.5

package/README.md

@@ -4,7 +4,7 @@
 
 [![npm](https://img.shields.io/npm/v/@path58/p58-n8n)](https://www.npmjs.com/package/@path58/p58-n8n)
 [![License: BSL 1.1](https://img.shields.io/badge/License-BSL_1.1-orange.svg)](LICENSE)
-[![Tests: 9,799 passed](https://img.shields.io/badge/Tests-9%2C799_passed-brightgreen)](https://github.com/tsvika58/p58-n8n/actions/workflows/ci.yml)
+[![Tests: 9,860 passed](https://img.shields.io/badge/Tests-9%2C860_passed-brightgreen)](https://github.com/tsvika58/p58-n8n/actions/workflows/ci.yml)
 [![Node](https://img.shields.io/badge/node-%3E%3D18-brightgreen)](https://nodejs.org)
 
 p58-n8n is an MCP server that gives your AI assistant deep knowledge of n8n — **1,545 nodes**, **12,619 operations**, **679 credential types** — so it can plan, build, validate, fix, and deploy workflows correctly.
@@ -221,6 +221,14 @@ Validation runs against real catalog data, not LLM inference. When issues are fo
 
 Structured catalog data means smaller prompts, fewer turns, and less token waste. On Haiku 4.5: **$0.24/workflow** vs $6.02 for the next-best server and $7.16 for no server at all.
 
+### ⭐ Premium Catalog Access
+
+The free public catalog covers 1,545 node names and basic metadata. **Premium access** unlocks the full enriched catalog — operations, parameters, credential schemas, config examples, connection rules — served via a dedicated REST API. No database credentials needed on your machine. Activate directly from your AI assistant:
+
+> *"Run `activate_premium` with my access token"*
+
+Your token is validated, stored locally at `~/.config/p58-n8n/config.json`, and used automatically on every session. If the premium API is unreachable, p58-n8n gracefully falls back to the public catalog — you keep working without interruption.
+
 ---
 
 ## Security & Credential Safety