Skip to content

Connect your client ​

Every client needs the same one thing:

text
https://mcp.zefi.ai/mcp

What differs is where you put it. If your client is not listed, Any other client covers it.

The flow is the same everywhere

Add the URL, the client opens your browser, you sign in to Zefi and choose a workspace. No key, no secret, no client id to request from us.

Claude ​

For Claude on the web and Claude Desktop.

  1. Open Settings → Connectors. On the web that is your profile icon then Settings; on the desktop app, Ctrl+, or ⌘,.
  2. Click Add custom connector.
  3. Paste https://mcp.zefi.ai/mcp and click Add.
  4. Click Connect on the new entry, sign in, and choose your workspace.

On a Team or Enterprise plan an admin may need to allow custom connectors first.

Claude Code ​

bash
claude mcp add --transport http zefi https://mcp.zefi.ai/mcp
json
{
  "mcpServers": {
    "zefi": {
      "type": "http",
      "url": "https://mcp.zefi.ai/mcp"
    }
  }
}

Adding it does not sign you in. Run /mcp inside a session, select zefi, and choose Authenticate.

The command adds the server for the current project. Add --scope user for every project, or --scope project to write the .mcp.json above into the working tree for the whole team to commit. That file carries the address, never a credential, so each person still signs in as themselves.

claude mcp logout zefi clears a sign-in.

ChatGPT ​

Your own MCP servers live behind developer mode, and ChatGPT now calls them plugins rather than connectors.

  1. Open Settings → Security and login and turn on Developer mode.
  2. Go to chatgpt.com/plugins and select the + button.
  3. Enter a name and description — both are required — then paste https://mcp.zefi.ai/mcp as the MCP server URL. The /mcp path is part of it.
  4. Create the connection, sign in, and choose your workspace.

Add the connection to a conversation from the tools menu.

If ChatGPT's settings look different

ChatGPT has renamed this area twice — connectors, then apps, then plugins — so older guides point at a Settings → Connectors page it no longer has. In ChatGPT, developer mode lives under Security and login. (Claude's Connectors page above is a different product and is current.)

Availability depends on your plan, and on business plans an admin may need to allow developer mode first.

Codex ​

bash
codex mcp add zefi --url https://mcp.zefi.ai/mcp
toml
# ~/.codex/config.toml
[mcp_servers.zefi]
url = "https://mcp.zefi.ai/mcp"

The command starts the sign-in for you, printing a URL to open. Follow it, sign in, and choose your workspace.

If the command is unrecognised, update the CLI — remote MCP support is recent — or use the bridge.

Cursor ​

Add Zefi to Cursor — one click, if you are reading this on the machine Cursor is installed on.

Otherwise add it by hand, in ~/.cursor/mcp.json for every project or .cursor/mcp.json inside one:

json
{
  "mcpServers": {
    "zefi": {
      "url": "https://mcp.zefi.ai/mcp"
    }
  }
}

Either way, open Settings → MCP, where Zefi now appears, and click it to sign in.

VS Code ​

For GitHub Copilot's agent mode.

bash
code --add-mcp '{"name":"zefi","type":"http","url":"https://mcp.zefi.ai/mcp"}'
json
{
  "servers": {
    "zefi": {
      "type": "http",
      "url": "https://mcp.zefi.ai/mcp"
    }
  }
}

For a guided version instead, run MCP: Add Server from the command palette and choose whether it applies to this workspace or to you everywhere.

Note the key is servers, not mcpServers — VS Code is the odd one out here. Open the Chat view in Agent mode and the server prompts you to sign in on first use. MCP: List Servers shows its status and lets you re-authenticate.

Antigravity ​

The Antigravity CLI and the Antigravity IDE share one configuration, so either of these covers both.

bash
agy mcp add zefi https://mcp.zefi.ai/mcp
json
// ~/.gemini/config/mcp_config.json, or .agents/mcp_config.json in a project
{
  "mcpServers": {
    "zefi": {
      "serverUrl": "https://mcp.zefi.ai/mcp"
    }
  }
}

Adding it does not sign you in. Type /mcp in the prompt panel, select zefi and authenticate.

In the IDE, that same file is behind … → MCP Servers → Manage MCP Servers → View raw config.

Three things that look similar and are not

Gemini CLI was retired on 18 June 2026 and replaced by Antigravity CLI. If you still have it, httpUrl was its key and that configuration does not carry over — Antigravity rejects url and httpUrl outright. Enterprise Gemini Code Assist licences keep the old CLI working, and there httpUrl is still correct.

The Gemini app at gemini.google.com cannot add a custom MCP server at all.

Gemini Enterprise has one, added by a team administrator under Settings & help → Team → Manage team → Connected apps, but it asks for an OAuth client ID and client secret. Zefi issues no client secret — MCP clients are public clients, where PKCE does the binding and a secret would only be something to leak — so that connector cannot complete its setup against this server today. Talk to us if you need it.

Windsurf ​

Now published as Devin Desktop, where Cascade is the agent.

Click the MCPs icon at the top right of the Cascade panel — or open Devin Settings → Cascade → MCP Servers — and add a custom server. To write it by hand instead, the file is ~/.codeium/windsurf/mcp_config.json and the key is serverUrl:

json
{
  "mcpServers": {
    "zefi": {
      "serverUrl": "https://mcp.zefi.ai/mcp"
    }
  }
}

Press Refresh in the MCP panel, then sign in when prompted.

Zed ​

Zed calls these context servers, and reaches remote ones directly.

Open Settings → AI → MCP Servers, choose Add Server → Add Remote Server, and give it the URL. Or write it into settings.json:

json
{
  "context_servers": {
    "zefi": {
      "url": "https://mcp.zefi.ai/mcp"
    }
  }
}

With no Authorization header configured, Zed runs the sign-in itself — there is nothing to install.

Any other client ​

Any client that can run a local MCP server can reach a remote one through mcp-remote, a small bridge that speaks stdio to your client and HTTP to us. It handles the OAuth flow the same way.

json
{
  "mcpServers": {
    "zefi": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.zefi.ai/mcp"]
    }
  }
}

It needs Node.js 18 or later. Prefer a client's native HTTP support where it exists — the bridge keeps its own copy of your sign-in on disk.

If a sign-in gets stuck, clearing ~/.mcp-auth and reconnecting resets it.

Checking it worked ​

Ask your assistant which Zefi tools it has, and to tell you which workspace it is connected to. A healthy connection answers both.

If the workspace is not the one you meant, disconnect and connect again — the workspace is fixed when you approve, so it cannot be changed in place.

Anything else, see Troubleshooting.

Zefi API v1