@path58/p58-n8n 0.2.0

package/AGENT_INSTALL.md

@@ -0,0 +1,223 @@
+# P58-n8n — Drop & Install
+
+**For users:** Drop this file into any AI chat and say: **"Install this MCP server for me"**
+
+**For AI agents:** Follow the instructions in the AI AGENT section below.
+
+---
+
+## What you'll get
+
+P58-n8n gives your AI assistant deep n8n workflow knowledge — 1,982 nodes, 40,179 operations, 26 auto-fix rules.
+
+| Tool | What it does |
+|------|-------------|
+| `validate_workflow` | Run L1-L6 validation on workflow JSON |
+| `get_operation_schema` | Get exact parameter requirements for any operation |
+| `check_parameter` | Validate a parameter value with typo detection |
+| `suggest_fix` | Get fix suggestions for validation issues |
+| `list_nodes` | Browse and search 1,982 n8n nodes |
+| `list_operations` | List all operations for a node type |
+| `get_node_info` | Get comprehensive node metadata |
+| `find_similar_pattern` | Search 9,000+ validated workflows |
+
+### Supported clients
+
+| Client | Support | Transport |
+|--------|---------|-----------|
+| Claude Desktop (Chat + Cowork) | Primary | stdio |
+| Claude Code (CLI) | Primary | stdio |
+| Gemini CLI | Secondary | stdio |
+| ChatGPT Desktop | Not supported | Requires HTTP/SSE |
+
+---
+
+## AI AGENT: Follow These Instructions
+
+### STEP 1 — Detect Environment
+
+Determine your capabilities and the user's platform.
+
+**Can you execute commands?**
+```
+Can I run shell commands?
+├─ YES (Claude Code, Cowork mode, Gemini CLI)  → Go to STEP 2a (Auto-Install)
+└─ NO  (Claude Desktop Chat-only)              → Go to STEP 2b (Guided Install)
+```
+
+**Detect platform:**
+```
+What OS is the user on?
+├─ macOS    → Config paths use ~/Library/Application Support/
+├─ Windows  → Config paths use %APPDATA%\
+└─ Linux    → Config paths use ~/.config/
+```
+
+---
+
+### STEP 2a — Auto-Install (can run commands)
+
+**Prerequisites — verify first:**
+
+```bash
+node --version
+# Must be v18.0.0 or higher. If not installed, tell the user to install from https://nodejs.org/
+```
+
+**Then install based on which client the user wants:**
+
+#### Claude Code
+
+Run this single command:
+
+```bash
+claude mcp add p58-n8n -- npx -y @path58/p58-n8n
+```
+
+Done. Skip to STEP 3.
+
+#### Claude Desktop (Cowork mode — can write files)
+
+Read the existing config file first, then MERGE the p58-n8n entry. NEVER overwrite the entire file.
+
+**Config file paths:**
+- macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
+- Windows: `%APPDATA%\Claude\claude_desktop_config.json`
+- Linux: `~/.config/Claude/claude_desktop_config.json`
+
+**If file exists:** Read it, parse the JSON, add this entry inside `mcpServers`:
+
+```json
+"p58-n8n": {
+  "command": "npx",
+  "args": ["-y", "@path58/p58-n8n"]
+}
+```
+
+**If file doesn't exist:** Create it with:
+
+```json
+{
+  "mcpServers": {
+    "p58-n8n": {
+      "command": "npx",
+      "args": ["-y", "@path58/p58-n8n"]
+    }
+  }
+}
+```
+
+Tell the user to restart Claude Desktop.
+
+#### Gemini CLI
+
+There is NO `gemini mcp add` command. Edit the config file directly.
+
+**Config path:** `~/.gemini/settings.json` (user-level) or `.gemini/settings.json` (project-level)
+
+**If file exists:** Read it, parse JSON, add the `p58-n8n` entry inside `mcpServers`.
+
+**If file doesn't exist:** Create it with:
+
+```json
+{
+  "mcpServers": {
+    "p58-n8n": {
+      "command": "npx",
+      "args": ["-y", "@path58/p58-n8n"]
+    }
+  }
+}
+```
+
+---
+
+### STEP 2b — Guided Install (chat-only)
+
+Ask the user which client they want to set up, then give them copy-paste instructions.
+
+**Ask:** "Which AI client do you want to use P58-n8n with?"
+
+#### For Claude Desktop
+
+Tell the user:
+
+1. Open your Claude Desktop config file:
+   - **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
+   - **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
+   - **Linux:** `~/.config/Claude/claude_desktop_config.json`
+
+2. Add this inside the `mcpServers` object (create the file if it doesn't exist):
+
+```json
+{
+  "mcpServers": {
+    "p58-n8n": {
+      "command": "npx",
+      "args": ["-y", "@path58/p58-n8n"]
+    }
+  }
+}
+```
+
+3. Restart Claude Desktop.
+
+#### For Claude Code
+
+Tell the user to run:
+
+```bash
+claude mcp add p58-n8n -- npx -y @path58/p58-n8n
+```
+
+#### For Gemini CLI
+
+Tell the user:
+
+1. Open `~/.gemini/settings.json` (create if needed)
+2. Add:
+
+```json
+{
+  "mcpServers": {
+    "p58-n8n": {
+      "command": "npx",
+      "args": ["-y", "@path58/p58-n8n"]
+    }
+  }
+}
+```
+
+#### For ChatGPT Desktop
+
+Tell the user: **ChatGPT Desktop is not supported.** It requires HTTP/SSE transport for MCP servers, but P58-n8n uses stdio transport. This may be supported in a future version.
+
+---
+
+### STEP 3 — Verify Installation
+
+After setup, test with this prompt:
+
+> "List available n8n nodes for email"
+
+**Expected result:** The AI calls `list_nodes` and returns nodes like `n8n-nodes-base.gmail`, `n8n-nodes-base.emailSend`, etc.
+
+**If it doesn't work:**
+
+```
+Tools not appearing?
+├─ Config file in wrong location   → Check OS-specific paths above
+├─ Invalid JSON syntax             → Validate with: cat <path> | python3 -m json.tool
+├─ Client not restarted            → Close and reopen the AI client
+├─ npx not found                   → Install Node.js 18+ from https://nodejs.org/
+└─ Permission error                → Try: sudo npm install -g @path58/p58-n8n
+
+Server starts but tools don't work?
+├─ Database not accessible         → Server needs network access to Supabase
+├─ Timeout errors                  → Check internet connection
+└─ "Module not found"              → Run: npm install -g @path58/p58-n8n (reinstall)
+```
+
+---
+
+**Package:** `@path58/p58-n8n` | **License:** MIT | **Built by [Path58](https://path58.com)**

package/CHANGELOG.md

@@ -0,0 +1,88 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.2.0] - 2026-03-08
+
+### Added
+
+- **26 new MCP tools** — total now 34 across 7 tiers (up from 8 in v0.1.0)
+  - **Tier 2 — Server CRUD (7 tools):** `get_workflow`, `list_workflows`, `create_workflow`, `update_workflow`, `delete_workflow`, `execute_workflow`, `get_execution_result`
+  - **Tier 3 — Advanced Operations (5 tools):** `validated_create`, `validated_update`, `activate_workflow`, `get_session_metrics`, `server_health`
+  - **Tier 4 — Tool Parity (5 tools):** `test_workflow`, `server_autofix`, `partial_update_workflow`, `manage_executions`, `deploy_template`
+  - **Tier 5 — Workflow Generation (2 tools):** `plan_workflow`, `build_workflow`
+  - **Tier 6 — Credential Management (6 tools):** `list_credentials`, `get_credential_schema`, `create_credential`, `test_credential`, `update_credential`, `delete_credential`
+  - **Tier 7 — Config Intelligence (1 tool):** `collect_config`
+- **Credential management** — full lifecycle: list, inspect schema (679 credential types), create, health-check, update, delete; 3-layer redaction prevents secrets in tool output
+- **Workflow generation pipeline** — `plan_workflow` + `build_workflow` produce validated workflows from natural language; multi-pass auto-fix runs automatically
+- **Config intelligence** — `collect_config` detects missing node configuration using 2,494 known gaps
+- **validated_create / validated_update** — L1-L6 validation runs before any deploy; rejects invalid JSON before it reaches the n8n server
+- **Diff-based updates** — `partial_update_workflow` applies node-level patches without replacing the entire workflow
+- **K2 sandbox testing** — `test_workflow` validates workflow logic using mock credential responses (no real API calls required)
+- **Session cost metrics** — every tool response includes token usage and cost; `get_session_metrics` returns cumulative session totals
+- **Auto-fixer expanded** — 35 production fixers (up from 26), including 5 new L6 V2 fixers for flow integrity and an LLM fallback fixer
+- **L5 safety guard** — prevents destructive auto-fixes on uncataloged community nodes
+- **Catalog refreshed (2026-03-08)** — nodes: 1,545 (1,482 core + 63 community), credentials: 679, operations: 12,619, parameters: 38,005
+- **Cursor and Windsurf support** — installation guide updated with setup instructions for both IDEs
+- **Benchmark results (RAG-4.36)** — 287-observation study: p58 90% vs n8n-mcp 35% vs FlowEngine 35% vs no-MCP 25%; $0.25/prompt vs $0.76 (67% cheaper)
+
+### Changed
+
+- README completely rewritten to reflect v0.2.0 capabilities, fresh catalog counts, and competitive positioning
+- TOOLS.md updated with all 34 tools, full input/output schemas, and configuration guide
+- INSTALLATION.md updated with Cursor, Windsurf, and multi-environment setup
+
+### Fixed
+
+- `partial_update_workflow` connection format bug — n8n expects `{ main: [[targets]] }` wrapper (was crashing on all models)
+- `server_autofix` JSON string input — now accepts workflow as JSON string or object
+- n8n rejects empty `pinData: {}` — removed from workflow payload when empty
+
+## [0.1.1] - 2026-02-06
+
+### Fixed
+
+- Bundle MCP server with esbuild to resolve private registry dependency (`@tsvika58/shared-utilities`) for `npx` installs
+- Set npm package access to public (scoped packages default to private)
+
+### Changed
+
+- `bin` entry now points to bundled CJS file (`dist/mcp/server.bundle.cjs`) for universal compatibility
+- Moved `@tsvika58/shared-utilities` from runtime to dev dependency (bundled into dist)
+
+## [0.1.0] - 2026-02-06
+
+### Added
+
+- **MCP Server** with 8 tools for LLM integration via stdio transport
+  - `validate_workflow` — Full L1-L6 static validation pipeline
+  - `get_operation_schema` — Parameter requirements lookup for any node operation
+  - `check_parameter` — Real-time parameter validation against catalog
+  - `suggest_fix` — Concrete fix suggestions powered by 26 deterministic auto-fixers
+  - `list_operations` — Browse all operations for a given node type
+  - `list_nodes` — Search and browse 1,982 available n8n nodes
+  - `get_node_info` — Full node details, configuration, and metadata
+  - `find_similar_pattern` — Working examples from 9,178 real workflows
+- **L1-L6 Static Validation Pipeline** (~1.4s total)
+  - L1: JSON structure, required fields, duplicate detection (~30ms)
+  - L2: Node type catalog validation (~60ms)
+  - L3: Credential type validation (~80ms)
+  - L4: Connection topology validation (~180ms)
+  - L5: Parameter configuration validation (~850ms)
+  - L6: Flow integrity, security, and expression patterns (~200ms)
+- **AutoFix Infrastructure** with 26 deterministic fixers across L1-L6
+  - Zero LLM cost for automatic repairs
+  - 97%+ success rate on fixable errors
+- **Node Catalog** with 1,982 nodes, 654 credential types, 40,179 operations
+- **Pattern Library** with 9,178 real workflow patterns for discovery
+- **Multi-client support** for Claude Desktop, Claude Code, and Gemini CLI
+- **Drop-and-install** configuration via `AGENT_INSTALL.md`
+- npm package published as `@path58/p58-n8n`
+- ESM module support with shebang for direct `npx` execution
+
+[0.2.0]: https://github.com/tsvika58/n8n-workflow-validator/releases/tag/v0.2.0
+[0.1.1]: https://github.com/tsvika58/n8n-workflow-validator/releases/tag/v0.1.1
+[0.1.0]: https://github.com/tsvika58/n8n-workflow-validator/releases/tag/v0.1.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Path58
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,281 @@
+# p58-n8n — The n8n MCP Server That Gets Workflows Right
+
+**Build n8n workflows from plain English. First try. Every time.**
+
+[![npm](https://img.shields.io/npm/v/@path58/p58-n8n)](https://www.npmjs.com/package/@path58/p58-n8n)
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
+[![Tests](https://github.com/tsvika58/n8n-workflow-validator/actions/workflows/ci.yml/badge.svg)](https://github.com/tsvika58/n8n-workflow-validator/actions/workflows/ci.yml)
+
+p58-n8n is an MCP server that gives your AI assistant deep knowledge of n8n — **1,545 nodes**, **12,619 operations**, **38,005 parameter rules** — so it can plan, build, validate, fix, and deploy workflows correctly. No manual iteration. No broken JSON.
+
+---
+
+## Quick Start (< 2 minutes)
+
+### 1. Install
+
+**Via npx (no install needed):**
+```bash
+npx -y @path58/p58-n8n
+```
+
+**Or install globally:**
+```bash
+npm install -g @path58/p58-n8n
+```
+
+### 2. Configure your AI client
+
+**Claude Code** (one command):
+```bash
+claude mcp add p58-n8n -- npx -y @path58/p58-n8n
+```
+
+**Claude Desktop** — add to `claude_desktop_config.json`:
+```json
+{
+  "mcpServers": {
+    "p58-n8n": {
+      "command": "npx",
+      "args": ["-y", "@path58/p58-n8n"]
+    }
+  }
+}
+```
+
+Config file location: `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows). Restart Claude Desktop after saving.
+
+**Cursor / Windsurf** — see [docs/INSTALLATION.md](docs/INSTALLATION.md) for full multi-client setup.
+
+### 3. Try it
+
+```
+Build me a workflow that monitors a Google Sheet and sends a Slack message
+when a new row is added. Use the Google Sheets and Slack credentials I already
+have configured in n8n.
+```
+
+p58-n8n will plan the workflow, look up the exact node parameters, validate the JSON against L1-L6 rules, wire your existing credentials, and deploy it to your n8n instance — in one conversation.
+
+---
+
+## What Makes p58 Different
+
+### Benchmark: 90% success rate on real-world workflows
+
+| | **p58-n8n** | **n8n-mcp** | **FlowEngine** | **Claude alone** |
+|---|---|---|---|---|
+| **Workflow success rate** | **90%** | 35% | 35% | 25% |
+| **Cost per prompt** | **$0.25** | $0.76 | — | ~$0.05 |
+| **Credential management** | ✅ Full CRUD | ❌ | ❌ | ❌ |
+| **Config intelligence** | ✅ 2,494 known gaps | ❌ | ❌ | ❌ |
+
+*Benchmarked on 287 real-world workflow generation tasks across Sonnet, Opus, and Haiku models.*
+
+### The full pipeline — not just JSON generation
+
+Most tools generate workflow JSON and stop. p58-n8n runs the complete cycle:
+
+```
+plan_workflow → build_workflow → validate (L1-L6) → server_autofix →
+wire credentials → collect_config → validated_create → execute_workflow
+```
+
+Every step is backed by structured catalog data. No guessing.
+
+### Deep validation — 6 levels, not vibes
+
+```
+L1 Structure  →  L2 Nodes  →  L3 Credentials  →  L4 Connections  →  L5 Parameters  →  L6 Patterns
+   ~30ms          ~60ms           ~80ms              ~180ms              ~850ms            ~200ms
+```
+
+L5 alone checks 38,005 parameter rules. L6 catches dead ends, retry loops without error handlers, and hardcoded secrets. Combined: ~1.4 seconds for a full audit.
+
+### 35 auto-fixers — not a prompt
+
+When validation fails, p58-n8n applies deterministic fixes: reconnect orphaned nodes, add missing credentials, patch wrong parameter types, rebuild broken flow paths. 97%+ success rate. Zero LLM cost.
+
+### Credential management — a category no competitor touches
+
+```
+list_credentials → get_credential_schema → create_credential →
+test_credential → update_credential → delete_credential
+```
+
+The AI can discover what credentials your n8n instance has, look up what fields any of 679 credential types need, create or update credentials, health-check them, and wire them into new workflows — all without leaving the conversation.
+
+### 67% cheaper than the alternative
+
+p58-n8n uses 80% less token context than n8n-mcp by serving structured catalog data instead of documentation blobs. At scale: $0.25/prompt vs $0.76 for czlonkowski.
+
+---
+
+## 34 Tools in 7 Tiers
+
+### Tier 1 — Validation & Analysis
+
+| Tool | What it does |
+|------|-------------|
+| `validate_workflow` | Full L1-L6 validation — structure, nodes, credentials, connections, parameters, flow patterns |
+| `get_operation_schema` | Exact parameter requirements for any node operation (ground truth, not LLM guesses) |
+| `check_parameter` | Real-time parameter validation — type checking, enum validation, typo detection |
+| `suggest_fix` | Concrete fix suggestions for validation issues — rename, add, remove, or replace |
+| `list_nodes` | Browse and search 1,545 n8n nodes by category or keyword |
+| `list_operations` | All operations for a node type, grouped by resource |
+| `get_node_info` | Full node details — operations, credentials, properties, version |
+| `find_similar_pattern` | Search 10,003 real workflows for working examples |
+
+### Tier 2 — Server CRUD
+
+| Tool | What it does |
+|------|-------------|
+| `get_workflow` | Fetch a workflow from your n8n instance by ID |
+| `list_workflows` | List all workflows with filtering and pagination |
+| `create_workflow` | Create a workflow on your n8n server |
+| `update_workflow` | Replace a workflow entirely |
+| `delete_workflow` | Delete a workflow (safe — checks for dependents) |
+| `execute_workflow` | Trigger workflow execution |
+| `get_execution_result` | Fetch the result of a completed execution |
+
+### Tier 3 — Advanced Operations
+
+| Tool | What it does |
+|------|-------------|
+| `validated_create` | Validate + deploy in one step — rejects invalid JSON before it hits your server |
+| `validated_update` | Validate + update in one step |
+| `activate_workflow` | Toggle workflow active/inactive state |
+| `get_session_metrics` | Token usage and cost tracking for the current session |
+| `server_health` | Check your n8n instance connectivity and status |
+
+### Tier 4 — Tool Parity
+
+| Tool | What it does |
+|------|-------------|
+| `test_workflow` | Sandbox-test a workflow using K2 mock responses (no real credentials needed) |
+| `server_autofix` | Run all 35 auto-fixers on a workflow and return the repaired JSON |
+| `partial_update_workflow` | Diff-based partial updates — change one node without replacing the whole workflow |
+| `manage_executions` | List, filter, and delete workflow execution history |
+| `deploy_template` | Deploy a pre-built template from the workflow library |
+
+### Tier 5 — Workflow Generation
+
+| Tool | What it does |
+|------|-------------|
+| `plan_workflow` | Generate a structured workflow plan from a natural language description |
+| `build_workflow` | Build a complete workflow JSON from a plan — uses catalog data for correct parameters |
+
+### Tier 6 — Credential Management
+
+| Tool | What it does |
+|------|-------------|
+| `list_credentials` | List n8n server credentials with optional schema enrichment |
+| `get_credential_schema` | Required and optional fields for any of 679 credential types (fuzzy match) |
+| `create_credential` | Create credentials with 5-step validation |
+| `test_credential` | Health-check probe — verifies connectivity for top credential types |
+| `update_credential` | Partial update — GET → merge → PATCH (preserves unmodified fields) |
+| `delete_credential` | Safe deletion with dependent workflow scan |
+
+### Tier 7 — Config Intelligence
+
+| Tool | What it does |
+|------|-------------|
+| `collect_config` | Detect missing node configuration from your live n8n instance (covers 2,494 known gaps) |
+
+Full reference with input/output schemas: [docs/TOOLS.md](docs/TOOLS.md)
+
+---
+
+## Examples
+
+### Build a Slack notification workflow
+
+```
+Plan and build me a workflow that triggers every day at 9am, queries a PostgreSQL
+database for orders placed in the last 24 hours, and posts a summary to the
+#daily-reports Slack channel.
+```
+
+p58-n8n will: call `plan_workflow` → `build_workflow` → `validate_workflow` (L1-L6) → `list_credentials` to find your Slack and PostgreSQL credentials → `validated_create` to deploy to n8n.
+
+### Fix a broken workflow
+
+```
+I imported this workflow from n8n.io but it has errors. Can you validate and fix it?
+[paste workflow JSON]
+```
+
+p58-n8n will: call `validate_workflow` → identify L1-L6 issues → call `server_autofix` to apply all 35 fixers → return the repaired workflow with a diff of what changed.
+
+### Set up credentials and deploy
+
+```
+I need to create a Slack credential and then use it in a new workflow.
+My Slack bot token is in my clipboard.
+```
+
+p58-n8n will: call `get_credential_schema` for Slack → `create_credential` with validation → `test_credential` to verify → wire it into the workflow → deploy.
+
+---
+
+## Installation
+
+See [docs/INSTALLATION.md](docs/INSTALLATION.md) for full setup instructions:
+
+- Claude Desktop (macOS, Windows, Linux)
+- Claude Code (one-command setup)
+- Cursor and Windsurf
+- Global install vs npx
+- Troubleshooting
+
+---
+
+## Benchmarks
+
+Benchmarked on 287 real-world workflow generation tasks using prompts from the n8n community. Tests ran across three models (Haiku 4.5, Sonnet 4.6, Opus 4.6) with four configurations (p58-n8n, n8n-mcp, FlowEngine, and no MCP server).
+
+| Metric | p58-n8n | n8n-mcp | FlowEngine | No MCP |
+|--------|---------|---------|------------|--------|
+| **Overall success rate** | **90%** | 35% | 35% | 25% |
+| **Sonnet 4.6** | 80% | — | — | — |
+| **Opus 4.6** | 80% | — | — | — |
+| **Haiku 4.5** | 60% | — | — | — |
+| **Cost per prompt** | $0.25 | $0.76 | — | ~$0.05 |
+| **Credential prompts** | **78%** | 0% | 0% | 0% |
+
+*"Credential prompts" = workflows requiring credential setup that competitors cannot attempt at all.*
+
+---
+
+## License
+
+MIT — see [LICENSE](LICENSE) for details.
+
+**Built by [Path58](https://path58.com)**
+
+---
+
+## Contributing
+
+p58-n8n is in **soft launch** (friends & family). Issues and feedback welcome:
+
+- **Bugs:** [GitHub Issues](https://github.com/tsvika58/n8n-workflow-validator/issues)
+- **Email:** tvagman@gmail.com
+
+---
+
+## Architecture
+
+p58-n8n runs as a local stdio MCP server. No cloud services required for basic use.
+
+```
+AI Client (Claude / Cursor / Windsurf)
+    ↕ MCP Protocol (stdio)
+p58-n8n Server
+    ├── Validation Engine (L1-L6, ~1.4s)
+    ├── AutoFix Engine (35 fixers, 97%+ success)
+    ├── Node Catalog (1,545 nodes, 12,619 ops, 38,005 params)
+    └── n8n API Client (deploy, execute, credential management)
+```
+
+**Prerequisites:** Node.js 18+ and a running n8n instance (for server CRUD and credential tools; validation and catalog tools work offline).