flexx logoflexx docs

Introduction to flexx

flexx connects a machine to AI agents over MCP and gives those agents durable memory. Start with Memory — no install required — add the CLI for remote access, and bring your own models for Flexx Agent.

What is flexx?

flexx is one platform for connecting machines to AI agents, in four pieces:

  • Remote — reach a machine from anywhere over MCP or SSH. No VPN, no port forwarding, zero inbound ports.
  • Router — one place for your models. Bring Anthropic, OpenAI, OpenRouter, Groq, or any OpenAI-compatible key and use them everywhere.
  • Memory — a shared knowledge graph of decisions, preferences, and project context that any MCP client can read and write.
  • Agent — a chat assistant that already has hands, models, and memory wired together.

You can use any piece on its own with the client you already have, or use all of them in Flexx Agent.

What each step unlocks

Nothing here is all-or-nothing. This is the order the platform unlocks — Memory works immediately, the CLI is what adds remote access, and your own keys are what power the Agent:

StepWhat you doWhat it unlocks
0Create a free accountMemory — works right away, no install
1Install the flexx CLI on a machine and connect itRemote — use that machine from anywhere
2Add your own provider API keysRouter + Agent models — your models, your budgets

Only here for memory?

You can skip the CLI entirely. A free account plus any MCP client gives you the memory tools (memory_remember, memory_recall, memory_search, and more). Remote tools — files, shell, screenshots, tunnels — only appear once a machine with the CLI is online. See Memory.

Quick start

Memory only — no install

  1. Create a free account at app.flexx.dev.
  2. Add https://relay.flexx.dev/v1/mcp as an MCP server in your AI tool.
  3. Your agent can now remember things — see Memory.

Full remote access

# 1. Install the CLI on the machine you want to reach (macOS / Linux)
curl -fsSL https://cli.flexx.dev/install.sh | sh

# 2. Sign in, register the machine, and print your MCP config
flexx setup

# 3. Bring it online — or `flexx service install` for always-on
flexx connect

That machine is now reachable from anywhere. Add the MCP endpoint to Cursor, Claude, ChatGPT, or any MCP client — see MCP configuration.

Your own models

Add a provider key in Dashboard → Models to power Flexx Agent and the router, or point your own code at the router's OpenAI- and Anthropic-compatible endpoints. See Bring your own models.

Key features

  • Remote machine access over MCP — files, shell, screenshots, tunnels, and code search on your machine
  • SSH access — ssh flexx-myserver works like any SSH host
  • Localhost tunneling — expose a dev server publicly with a single command
  • Bring your own models — Anthropic, OpenAI, OpenRouter, Groq, or any OpenAI-compatible endpoint, with per-key budgets
  • Graph-backed memory — decisions, preferences, and project context, scoped per repository
  • Flexx Agent — a unified assistant with hands, models, and memory
  • Workspace safety — path scoping, command blocklist, file size limits
  • Zero inbound ports — outbound WebSocket only
  • Cross-platform — macOS, Linux, Windows (amd64 and arm64)

Supported AI tools

flexx works with any MCP-compatible client, including:

  • Cursor — add the MCP server URL in Settings → MCP
  • Claude Desktop — add to ~/.claude/mcp_servers.json
  • ChatGPT — use the MCP endpoint with OAuth
  • OpenAI Codex — standard MCP configuration
  • Any MCP client — standard Model Context Protocol endpoint

Next steps

On this page