Does n8n instance-level MCP work with Claude Code?
Yes. n8n instance-level MCP passed every step of our test in Claude Code on Oct 16, 2026: it installed, connected, its tools were visible, and an agent used it to finish a real task.
Result
Steps
- install OK
- connect OK
- tools visible OK
- agent task: node lookup OKread-only node catalog search (n8n node library), no account or workflow run
- agent task: run fixture OKdiscover + run an automation on a local, self-hosted instance: the fixture echoes its input, so the fresh {TOKEN} can only come back if the server actually ran it
From our own harness run, not the maker’s claims. How we test.
Versions tested
Claude Code 2.1.296 · server reported v1.2.0 · connect time not recorded · tested Oct 16, 2026
Server package and resolved dependency versions: not recorded for these runs (our harness records them from Oct 19, 2026 on).
What happened in this run
Install config we used
claude mcp add -s local n8n-instance-mcp -- npx -y supergateway --streamableHttp http://localhost:5678/mcp-server/http --header Authorization: Bearer <YOUR_N8N_MCP_TOKEN>Connection check
n8n-instance-mcp: npx -y supergateway --streamableHttp http://localhost:5678/mcp-server/http --header Authorization: Bearer <YOUR_N8N_MCP_TOKEN> - ✔ ConnectedTools check
verified via task step: agent saw and called the server's tools
Agent task: node lookup
- Tool calls, in order:
mcp__n8n-instance-mcp__search_nodes
Agent task: run fixture
- Tool calls, in order:
mcp__n8n-instance-mcp__search_workflows,mcp__n8n-instance-mcp__get_workflow_details,mcp__n8n-instance-mcp__execute_workflow,mcp__n8n-instance-mcp__get_workflow_execution
Raw details from our harness log for this run. Secrets redacted.
Set it up in Claude Code
Copy-paste config for every app is on the n8n instance-level MCP page. Not connecting? Claude Code MCP troubleshooting.
Other apps we tested it in: Cursor (passed).