Provides integration with an n8n instance via its REST API v1, enabling AI agents to manage workflows, nodes, executions, templates, credentials, snapshots, validation, diagnostics, and self-healing operations.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@n8n-mcp patch the HTTP Request node in workflow 12 to use POST and save a snapshot"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
n8n-mcp)
A production-grade, standalone Model Context Protocol (MCP) Server for n8n workflow automation. Built with clean architecture, strict Pydantic v2 domain schemas, Windows Proactor compatibility, and anti-hallucination engines for Claude Desktop, Cursor, Antigravity, and autonomous agent swarms.
📖 Looking for Prompting & Operational Recipes?
Check out the comprehensive PROMPTING_GUIDE.md (Bangla & English user manual) and PLAYBOOK.md for 10+ copy-paste prompts and self-healing recipes!
LLMs interacting with the raw n8n REST API struggle with several major bottlenecks:
Massive Token Bloat: Sending full 50-node workflow JSONs on every modification consumes 15,000–30,000+ tokens per turn.
HTTP 400 Bad Request on Updates: n8n rejects PUT requests containing read-only fields (id, versionId, createdAt, updatedAt, triggerCount, tags).
Hallucinated Node Parameters: AI models regularly hallucinate non-existent properties, invalid credentials, or outdated enum values.
Fragile LangChain / AI Agent Sub-graphs: Missing required ai_languageModel or tool connections crashes AI nodes at runtime.
No Built-in Safety Net: Modifying live workflows without automated snapshotting and 1-click rollback risks production downtime.
n8n-mcp solves all of these problems natively.
Related MCP server: n8n MCP Agent
graph TD
Client[AI Client / Claude Desktop / Cursor / Antigravity]
subgraph MCP Server Layer ["n8n MCP Server (FastMCP / Stdio)"]
Transport[Stdio Transport / Windows Proactor Loop]
Registry[Central Tool Registry - 26 Production Tools]
end
subgraph Domain Engines ["Domain Engines & Core Logic"]
Patcher[Diff-Based Node Patcher & Deep-Merge]
Snapshots[Timestamped Snapshot & 1-Click Rollback]
PinData[Pin-Data Sandbox Testing Engine]
Catalog[Anti-Hallucination Node Catalog]
Validator[DAG Cycle & LangChain Multi-Port Linter]
Diagnostics[Execution Diagnostics & RCA Engine]
Healer[Autonomous Self-Healing Loop max 2 attempts]
Stealth[behavioral-playwright Anti-Bot Bridge]
ClientCore[Resilient REST Client Full-Jitter Backoff]
end
subgraph External Infrastructure ["External Infrastructure"]
N8nAPI[n8n Instance REST API v1]
PlaywrightEngine[behavioral-playwright Stealth Engine]
TemplateLib[n8n Official & Community Templates]
end
Client <-->|JSON-RPC 2.0 / Stdio| Transport
Transport --> Registry
Registry --> Patcher
Registry --> Snapshots
Registry --> PinData
Registry --> Catalog
Registry --> Validator
Registry --> Diagnostics
Registry --> Healer
Registry --> Stealth
Registry --> ClientCore
Patcher --> Snapshots
Patcher --> ClientCore
Healer --> Diagnostics
Healer --> Patcher
Healer --> ClientCore
PinData --> ClientCore
Stealth --> PlaywrightEngine
ClientCore <--> N8nAPI
ClientCore <--> TemplateLib# | Tool Name | Category | Description | Token Advantage |
1 |
| Workflows | List all workflows with active status, tags, and cursor pagination | Filtered minimal DTO |
2 |
| Workflows | Retrieve full workflow JSON (nodes, connections, settings, pinData) | Read-only inspection |
3 |
| Workflows | Create new workflow with auto-sanitized payload | Strips read-only fields |
4 |
| Workflows | Replace workflow definition with guaranteed HTTP 400 prevention | Zero read-only rejection |
5 |
| Diff Patcher | Partial node patcher: In-memory deep-merge of parameters with pre-patch snapshot | 85–90% token reduction |
6 |
| Snapshots | 1-click rollback: Restores previous workflow state from local snapshots | Instant recovery |
7 |
| Workflows | Activate or deactivate a workflow in n8n | Atomic boolean toggle |
8 |
| Workflows | Permanently deletes a workflow by ID | Direct cleanup |
9 |
| Discovery | Zero-latency keyword search across core & AI nodes | Eliminates hallucination |
10 |
| Discovery | Inspect parameter contract, enums, and required credentials for any node | Exact specification |
11 |
| Validation | Multi-port DAG cycle detection, dangling nodes check, expression syntax linter, and Python AST | Pre-deployment sanity gate |
12 |
| AI Validation | Validates LangChain agent sub-nodes ( | Guarantees AI graph correctness |
13 |
| Testing | Injects mock test data into trigger nodes without running external webhooks | Safe sandbox testing |
14 |
| Testing | Clears pinned data from specific or all nodes prior to live deployment | Clean production release |
15 |
| Executions | Query execution history by workflow, status ( | Minimal execution DTO |
16 |
| Executions | Detailed execution inspection with full step-by-step I/O and runtime data | Deep debugging |
17 |
| Diagnostics | Pinpoints crashed node and produces structured Root Cause Analysis (RCA) | Instant error diagnosis |
18 |
| Executions | Retries failed execution with optional latest workflow definition reloading | Resilient recovery |
19 |
| Executions | Purges execution records from n8n history | History pruning |
20 |
| Self-Healing | Autonomous loop: Diagnoses RCA -> patches node -> retries execution (max 2 attempts) | Zero human intervention |
21 |
| Templates | Searches 2,000+ official and community workflow templates by use case | Rapid scaffolding |
22 |
| Templates | Downloads verified workflow template JSON ready for deployment | Instant template cloning |
23 |
| Stealth Bridge | Generates | Cloudflare/Turnstile bypass |
24 |
| Operations | Dispatches test ( | Dynamic manual trigger |
25 |
| Operations | Pings n8n public API to verify connectivity, latency, and instance status | Liveness probe |
26 |
| Security | Lists credential IDs and types with secret values safely masked | Leak-proof discovery |
Traditional n8n MCP servers force the LLM to read and rewrite the entire workflow JSON on every edit:
Metric | Traditional MCP Server |
| Improvement |
Single Parameter Change | ~22,000 tokens (Full JSON roundtrip) | ~180 tokens ( | 99.2% Savings |
Multi-node Update | ~35,000 tokens | ~850 tokens | 97.5% Savings |
Context Window Saturation | Reaches limit in 3–4 edits | Stays under 5% over 50+ edits | 10x Longer Sessions |
Safety & Rollback | Manual undo or lost state | Automated snapshot before patch | Zero Data Loss |
git clone https://github.com/sadik004/n8n.mcp.git
cd n8n.mcp
pip install -e .Copy .env.example to .env:
N8N_HOST=http://localhost:5678
N8N_API_KEY=your_n8n_public_api_key_here
TIMEOUT_SECONDS=30.0
MAX_RETRIES=3
BEHAVIORAL_PLAYWRIGHT_URL=http://host.docker.internal:8000
SNAPSHOTS_DIR=.snapshotspython -m n8n_mcp --checkclaude_desktop_config.json){
"mcpServers": {
"n8n": {
"command": "python",
"args": ["-m", "n8n_mcp", "--transport", "stdio"],
"env": {
"N8N_HOST": "http://localhost:5678",
"N8N_API_KEY": "YOUR_N8N_API_KEY",
"BEHAVIORAL_PLAYWRIGHT_URL": "http://host.docker.internal:8000"
}
}
}
}mcp.json){
"mcpServers": {
"n8n": {
"command": "n8n-mcp",
"args": ["--transport", "stdio"],
"env": {
"N8N_HOST": "http://localhost:5678",
"N8N_API_KEY": "YOUR_N8N_API_KEY"
}
}
}
}Every layer is rigorously covered with unit and end-to-end integration tests:
pytest -v============================= test session starts =============================
platform win32 -- Python 3.13.9, pytest-8.4.2, pluggy-1.5.0
collected 54 items
tests/integration/test_mcp_e2e.py::test_fastmcp_initialization_and_tool_count PASSED [ 1%]
tests/integration/test_mcp_e2e.py::test_jsonrpc_initialize_handshake PASSED [ 3%]
...
tests/unit/test_validator.py::test_validate_ai_agent_graph_missing_language_model PASSED [ 98%]
tests/unit/test_validator.py::test_validate_ai_agent_graph_valid PASSED [100%]
============================= 54 passed in 3.01s ==============================This project is licensed under the MIT License.
This server cannot be deployed
Create, browse, remix, collaborate on, and run durable AI workflow nodes from MCP hosts.
Open-source Zapier/n8n alternative as an MCP server: agents build, run and debug your workflows.
Automate 1,000+ services from any MCP-compatible AI agent: build Applets, run actions and queries.
Zero-setup MCP gateway securely connecting AI to your tools with authentication and workflows
