MCP Configuration for AI Agents

Configure Cursor, Claude Desktop, ChatGPT, OpenAI Codex, and other MCP-compatible AI tools to connect to your machine via flexx. Includes config snippets and the complete list of available MCP tools.

What is MCP?

Model Context Protocol is an open standard for AI agents to interact with external tools. flexx exposes your machine's capabilities as MCP tools that any compatible agent can call.

Your MCP endpoint

After running flexx setup, your MCP endpoint is:

https://relay.flexx.dev/v1/mcp

Authentication is handled via OAuth 2.1 with PKCE — your AI tool will prompt you to approve access through a consent flow in the dashboard.

Configure your AI tool

Cursor

Open Settings → MCP → Add Server
Set the URL to:

https://relay.flexx.dev/v1/mcp

Cursor handles OAuth authentication automatically. You'll be prompted to approve access on first connection.

Claude Desktop

Add to ~/.claude/mcp_servers.json:

{
  "flexx": {
    "url": "https://relay.flexx.dev/v1/mcp"
  }
}

ChatGPT

Use the MCP endpoint URL directly in your ChatGPT MCP tool configuration:

https://relay.flexx.dev/v1/mcp

OAuth 2.1 with PKCE is supported — ChatGPT will handle the consent flow automatically.

OpenAI Codex

Add flexx as an MCP server in your Codex configuration:

https://relay.flexx.dev/v1/mcp

Any MCP client

flexx works with any tool that supports the Model Context Protocol standard:

Setting	Value
Endpoint URL	`https://relay.flexx.dev/v1/mcp`
Protocol	MCP (Model Context Protocol)
Authentication	OAuth 2.1 with PKCE
Transport	HTTP + Server-Sent Events

Available MCP tools

Tools are auto-discovered based on what's installed on your machine. Here is the complete list:

File operations

Tool	Description
`read_file`	Read file contents (supports offset and limit)
`write_file`	Write or create files
`list_directory`	List files and subdirectories
`delete_file`	Delete a file
`glob_files`	Find files matching glob patterns (e.g., `*/.ts`)

Code editing

Tool	Description
`str_replace`	Find and replace text in files (exact string match)

Shell

Tool	Description
`execute_command`	Run shell commands with configurable timeout. Supports optional `cwd` parameter to run from a subdirectory.

Search

Tool	Description
`grep_files`	Search file contents using ripgrep (requires ripgrep installed)

Screenshots

Tool	Description
`screenshot_url`	Capture a full-page screenshot of a URL (requires Chrome/Chromium)
`screenshot_element`	Capture a specific DOM element by CSS selector

Tunneling

Tool	Description
`tunnel_start`	Expose a local port via a public URL through the relay
`tunnel_stop`	Stop an active tunnel
`tunnel_list`	List all active tunnels

System

Tool	Description
`health_check`	Check daemon health and connectivity

Authentication flow

AI Agent
  │
  ├─ Connects to relay.flexx.dev/v1/mcp
  ├─ OAuth 2.1 + PKCE consent flow
  ├─ User approves access and selects remote
  ▼
Relay (relay.flexx.dev)
  │
  ├─ Validates OAuth token
  ├─ Routes to selected remote
  ▼
Your Machine (flexx daemon)
  │
  ├─ Executes MCP tool
  ├─ Returns result
  ▼
Relay → AI Agent

During the consent flow, you choose which remote machine the AI agent connects to. You can revoke access at any time from Dashboard → Settings.

OAuth scopes

Scope	Permission
`remotes:read`	View your connected machines and their status
`remotes:write`	Register, update, and remove machines
`tools:execute`	Run MCP tools on your machines
`profile:read`	View your account email and organization

Security

All communication is encrypted (WSS/HTTPS)
Your machine only makes outbound connections
OAuth 2.1 with PKCE prevents token interception
Tools respect workspace path scoping
Commands are filtered through a blocklist
File operations have size limits

MCP Configuration for AI Agents

On this page