MCP server
Ship n8n workflows from your editor.
One endpoint for Claude, Claude Code, Cursor, Windsurf, or any Streamable HTTP client. Search nodes, build workflows, and diagnose failures without leaving the window you are in.
Works in the clients you already use.
- Claude
- Claude Code
- Cursor
- Windsurf
Every tool says what it can touch.
The tools call the same API the web app does, so workspace scoping, the allowlisted patch schema, and the approval gate on native changes all apply when the caller is a model.
Read only
Search the node catalog, read workflows and patterns, validate a workflow. Safe to let an agent call freely.
search_nodesget_nodeget_workflow_patternsget_best_practiceslist_workflowsget_workflown8n_validate_workflow
Writes, labelled as such
Create, update, and adopt workflows. Annotated in the protocol so your client can ask before each call.
n8n_create_workflown8n_update_workflowadopt_n8n_workflow
Touches production
Run a workflow to diagnose it, heal it, restore a snapshot, or write native changes. Flagged destructive.
n8n_execute_and_diagnosen8n_heal_workflown8n_watch_workflowupdate_native_workflowrestore_workflow_snapshot
Connect in one command.
Issue a token in the workspace under Settings, then add the server to your client. Your default n8n connection is used unless you point a request somewhere else.
Claude CodeTerminal
claude mcp add --transport http --scope user pulsernode https://pulsernode.com/mcp \
--header "Authorization: Bearer pn_mcp_YOUR_TOKEN" \
--header "X-N8n-Url: https://your-n8n.example.com" \
--header "X-N8n-Key: YOUR_N8N_API_KEY"Cursor, Windsurf, Claude Desktopmcp.json
{
"mcpServers": {
"pulsernode": {
"type": "http",
"url": "https://pulsernode.com/mcp",
"headers": {
"Authorization": "Bearer pn_mcp_YOUR_TOKEN",
"X-N8n-Url": "https://your-n8n.example.com",
"X-N8n-Key": "YOUR_N8N_API_KEY"
}
}
}
}MCP questions, answered.
If your client speaks Streamable HTTP MCP, it works. Here is what it can do once connected.
Not here? Email hello@pulsernode.com
Which MCP clients can connect?
Any client that speaks Streamable HTTP MCP. That includes Claude, Claude Code, Cursor, and Windsurf. The server implements the official Model Context Protocol SDK, so a new compliant client works without changes on our side.
What can the MCP tools do?
Read-only tools search the node catalog, fetch node details and workflow patterns, list and read workflows, and validate a workflow. Write tools create, update, and adopt workflows. A third group is flagged destructive because it touches production directly: executing a workflow to diagnose it, healing it, restoring a snapshot, and writing native changes. Every tool carries these annotations in the protocol, so your client can prompt you before the risky ones run.
Is the MCP server safe to point an autonomous agent at?
The tools call the same API boundary the web app uses, so workspace scoping, the allowlisted patch schema, and approval gates on native changes all apply. An agent cannot reach your n8n instance by any path that skips those checks.
Is MCP access included on the Basic plan?
Yes. MCP access is part of Basic, along with version history, validation, and rollback. The read-only tools and the workflow create and update tools work on Basic; the AI workflow builder and Guard's healing tools require AI Pro.
The automation you keep putting off ships tonight.
Connect your n8n, describe the first one, and get back to your actual work.