Using Webflow’s MCP with Multiple Clients & Workspaces

Do you need to use Webflow's MCP server with multiple client sites- without continually disconnecting and reconnecting?

Here's the approach I use, which-

• Allows me to separate client MCP connections cleanly
• Authenticate ONCE for each
• Easily pick up where I left off
• Support advanced MCP configs, e.g. for site migrations or complex data loads

Setup

Define your MCP Config

Create your MCP config with the Claude CLI-

claude mcp add --transport http --scope project webflow-project1 https://mcp.webflow.com/mcp

Tips;

• Run this in the directory you will be opening in VS Code, which represents your specific client or project
• Remember to give your project a custom name, like webflow-project1
• Note the scoping with -s project , very important

Authenticate

Once the MCP is created, you need to authenticate.

✅ If Claude Code was already open, reload VSCode first so that it can see your new MCP config. CTRL+SHIFT+P and then Developer Reload.

If you are creating a new site, create that site in Webflow now. You can use a blank page if you're starting from scratch.

In the Claude Code extension or CLI, type /mcp

Click it, it will say "Needs Auth"

Click Authenticate

Approve and alerts that appear, to open the auth window

Complete Webflow's auth process

Primary Use Cases

• Easily switching between Claude MCP configurations, per client
• Ability to work simultaneously on multiple client projects without conflict, while still keeping their MCP access distinct

Advanced Use Cases

Site migrations

• Multiple MCP connections, possibly across workspaces
• Ability to protect a source prod site by restricting tools to read-only tools

More Research

MCP Gateways

Challenges

Claude code has a bug regarding token persistence, which requires some further research to minimize re-auth.

Notes

✅ Beginning Apr 2026, Anthropic began applying token limits to MCP responses at the harness level. To my knowledge this is limited to Claude Code, but relevant to this article.

I'm unsure how widespread this is, but this month Claude Code implemented a ~25K token limit on MCP responses. 🤔 I assume it's in response to the token consumption issues Anthropic users are hitting- but it means;

• Any WF MCP CMS call that involves rich text in the response is likely to fail. Even with only a few pgphs per field, Claude's having to paginate to ~20 items at a time to avoid the harness errors. Otherwise it gets downloaded into a temp directory and then Claude has to write python to try to extract it. It's... messy.
• There might be impacts to the designer API's on large pages.

Just surfacing this. The problem scenario I ran into today was a simple "list the 83 items in this collection."

Mitigation

You can override this limit;

.claude/settings.json

{
  "$schema": "https://json.schemastore.org/claude-code-settings.json",
  "env": {
    "MAX_MCP_OUTPUT_TOKENS": "200000"
  }
}