← All guides · Updated 2026-10-06

Connect Claude, Cursor or VS Code to W flow with MCP and let the AI build your workflows

W flow is an MCP server: connect Claude Code, Claude Desktop, Cursor or VS Code in two minutes, run your workflows as AI tools and let the assistant build new ones for you.

W flow speaks the Model Context Protocol (MCP). Connect your AI assistant once and it can do two things: run your workflows as tools (“send the invoice for ACME”) and build new workflows for you (“every hour, post new RSS items to Slack”) — which you then see, check and edit on the canvas.

What you need

• A free W flow account.
• One of: Claude Code, Claude Desktop, Cursor or VS Code (Copilot agent mode).

Step 1 — Create a token

1. In W flow open Settings → AI tools → Create token.
2. Give it a name (one per assistant or machine) and pick an access level: Run (only workflows you offer), Read (also look up nodes and workflows) or Build (also create, change and run workflows).
3. Copy the token. It starts with wfmcp_ and is shown only once.

The Settings page also shows a ready-to-paste config and the Claude Code command with your token already filled in.

Step 2 — Add W flow to your assistant

Claude Code — one command:

claude mcp add --transport http w-flow https://w-flow.tech/mcp \
  --header "Authorization: Bearer wfmcp_YOUR_TOKEN"

Check it with /mcp inside a session — W flow should show as connected.

Cursor — ~/.cursor/mcp.json:

{
  "mcpServers": {
    "w-flow": {
      "url": "https://w-flow.tech/mcp",
      "headers": { "Authorization": "Bearer wfmcp_YOUR_TOKEN" }
    }
  }
}

VS Code — .vscode/mcp.json:

{
  "servers": {
    "w-flow": {
      "type": "http",
      "url": "https://w-flow.tech/mcp",
      "headers": { "Authorization": "Bearer wfmcp_YOUR_TOKEN" }
    }
  }
}

Claude Desktop — it starts local programs, so bridge with mcp-remote (needs Node.js). Settings → Developer → Edit Config, then restart:

{
  "mcpServers": {
    "w-flow": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://w-flow.tech/mcp",
               "--header", "Authorization:${WFLOW_AUTH}"],
      "env": { "WFLOW_AUTH": "Bearer wfmcp_YOUR_TOKEN" }
    }
  }
}

Step 3 — Ask

With a Build token, try:

Use the W flow tools to build a workflow that checks https://example.com/feed.xml every hour and posts new items to my Slack channel #news. Validate it before saving and give me the editor link.

The assistant reads the W flow reference, looks up the right node types, validates the workflow and saves it to your account. Open the editor link and the finished workflow is on the canvas — every node visible and editable.

Offer your own workflows as tools

Open a workflow's settings (gear) and switch on Offer this workflow as a tool to AI assistants. Describe what it does and which inputs it needs — the assistant reads that to decide when to call it. Reconnect the assistant (in Claude Code: /mcp) to see the new tool.

Stay in control

• Limit a token to chosen workflows and set a runs per day limit.
• Every call appears in Executions as AI tool, with the token name, arguments and answer.
• Only a hash of the token is stored; revoking it takes effect immediately.
• Give a chat assistant a Run token and a coding agent Build access only where it needs it — a workflow can send mail or post messages.

Troubleshooting

• The client wants an OAuth login — the Authorization header is missing; add the server again with the header.
• Only wflow_* tools, none of my workflows — no workflow is offered as a tool yet, or the token's scope excludes it. Reconnect after changing it.
• The answer says running — long runs carry on; the assistant fetches the result with wflow_get_run.