Hands, brains, and memory for AI agents.
Install one CLI on a machine and use that PC from anywhere. Bring your own model keys. Keep context that lasts. Memory works with any account — the CLI is what unlocks remote access.
A free account unlocks Memory instantly. Install the CLI to reach a machine. Add your keys to power the Agent. Read the docs →
Get started
Three steps. Start wherever you are.
Each step unlocks a different part of flexx. You don't need all three to begin.
Memory works right now. Create a free account, connect any MCP client, and the memory tools are available — no install, no machine required. Remote tools only appear once the CLI is connected.
Start with MemoryFour products, one system
Use them together in Flexx Agent, or one at a time
Every piece stands on its own as an MCP server for the tools you already use. Together, they power Flexx Agent.
Remote
Give any AI real hands on your machine
MCP server for Cursor, Claude, Grok, or any client. AI agents run code, edit files, and execute commands on your machine — zero inbound ports.
Needs
Install the CLI on the machine you want to reach
Router
One place for every model
Bring your own models and providers, set budgets, and use a single OpenAI-compatible endpoint from anywhere — including inside Flexx Agent.
Needs
Add your own provider API keys
Memory
Agents that remember across tools and sessions
A shared knowledge graph over MCP. Store decisions, preferences, and project context once — recall it from any client or from Flexx Agent.
Needs
Nothing — works with any account, no install
Agent
The assistant that uses all three
A unified assistant that already has hands, a model layer, and durable memory — chat, and it acts, remembers, and picks the right model for the job.
Needs
A model — yours, or bring your own keys
Ways to use flexx
Start with what you have today
Architecture
Secure by design. Zero open ports.
Your machine connects outbound to the relay. AI agents connect to the relay. They never talk directly to your machine.
Your Machine
flexx connect
Relay
relay.flexx.dev
AI Agent
MCP / SSH
Your Machine
flexx connect
Relay
relay.flexx.dev
AI Agent
MCP / SSH
Outbound WebSocket only — zero inbound ports required on your machine
How it works
Reach a machine in under two minutes
Install, register, connect. After this, that PC is usable from anywhere. No network configuration required.
Router · bring your own models
Your keys. Your models. One endpoint.
Add a provider key in the dashboard and every model it can reach shows up in one place. Flexx Agent uses them, and your own tools can call the router too. Provider keys stay server-side — they never touch your client.
- Anthropic
- OpenAI
- OpenRouter
- Groq
- Any OpenAI-compatible endpoint
# In Flexx Agent — pick a model, nothing else to wire up.
# Every registered model appears in the session model picker.
# From your own code — point any OpenAI *or* Anthropic
# client at your router endpoint (Dashboard → Models):
OPENAI_BASE_URL=https://<your-router>/v1
OPENAI_API_KEY=flexxr_... # a router key you create
# /v1/chat/completions → OpenAI wire format
# /v1/messages → Anthropic wire format
# Per-key budgets + allowlists are enforced before
# any upstream call.Features
Everything your AI agent needs
Standard protocols, secure defaults, and the tools your agents actually need.
MCP Protocol
AI agents interact with your machine via standard MCP tools — file ops, shell commands, screenshots, and more.
SSH Access
ssh flexx-myserver — works like any SSH host. Auto-configures ProxyCommand and known hosts.
Localhost Tunneling
Expose dev servers publicly with a single command. No ngrok, no configuration needed.
Screenshots & Vision
AI agents capture webpage screenshots for visual validation and UI testing workflows.
Workspace Safety
Path scoping, command blocklist, and file size limits keep your machine secure by default.
Auto-discovery
flexx detects installed tools automatically: ripgrep, Docker, Jupyter, Chrome, GPUs, and more.
Agent memory
Graph-backed memory for preferences, decisions, and project knowledge. Semantic search and structured recall over MCP — no prompt stuffing.
Use cases
Built for how you actually work
Pricing
Free to start. No credit card.
Run flexx on your own machine at no cost. Connect your development environment to AI agents in under 60 seconds.
Open Source
Self-host flexx on your own infrastructure with complete privacy.
- Self-hosted relay & daemon
- Unlimited MCP tools
- Localhost tunneling
- Community support
Cloud Free
Hosted relay and zero-configuration OAuth for individual developers.
- Unlimited remotes
- Knowledge graph memory
- Encrypted reverse tunnels
- Audit logs & Vision
Team
For teams collaborating with shared remote development environments.
- Multi-user organization access
- Custom tunnel domains
- SSO integration
- Dedicated support
MCP Configuration
Works with every MCP-compatible AI tool
One MCP endpoint gives your AI tool Remote and Memory. Memory tools work as soon as you connect; Remote tools appear once a machine with the CLI is online. Toggle which capabilities are exposed from your dashboard at any time.
Add to your Cursor MCP settings (Settings → MCP → Add Server)
{
"mcpServers": {
"flexx": {
"url": "https://relay.flexx.dev/v1/mcp"
}
}
}FAQ
Frequently asked questions
Can I use flexx without installing anything?
Yes — for Memory. Create a free account and connect any MCP client; the memory tools need no install and no machine. Remote tools (files, shell, screenshots, tunnels) only become available after you install the flexx CLI on a machine and connect it. Flexx Agent can chat without a machine, but it needs a connected machine to act on one.
What do I need to get started?
Just a free account to use Memory. To use a machine remotely, install the flexx CLI on it (one command), then run flexx setup and flexx connect — that machine is then reachable from anywhere. To power Flexx Agent with your own models, add a provider API key in the dashboard.
What is MCP and how does flexx use it?
MCP (Model Context Protocol) is an open standard that lets AI agents interact with external tools. flexx turns your machine into an MCP server — AI agents like Cursor, Claude, and ChatGPT can read files, run commands, and edit code on your machine through the relay.
Is my machine exposed to the internet?
No. Your machine only makes outbound WebSocket connections to the relay. Zero inbound ports are opened. The relay mediates all communication between AI agents and your machine.
Which AI tools are supported?
Any tool that supports the Model Context Protocol (MCP) works with flexx. This includes Cursor, Claude Desktop, ChatGPT with MCP plugins, OpenAI Codex, and any custom MCP client.
How is authentication handled?
flexx uses OAuth 2.1 with PKCE for MCP connections. When an AI tool connects, it goes through a consent flow where you approve access and choose which machine to connect. You can revoke access anytime from the dashboard.
What tools can AI agents use on my machine?
AI agents can read and write files, execute shell commands, search codebases (via ripgrep), take screenshots (if Chrome is installed), list processes, and more. Available tools are auto-discovered based on what's installed.
Does it work behind a corporate firewall or NAT?
Yes. Because flexx only makes outbound connections, it works behind firewalls, NAT, and most corporate networks without any configuration. No VPN or port forwarding needed.
Is flexx free?
Yes. The free tier includes unlimited remotes, unlimited MCP tools, SSH access, and localhost tunneling. Managed and team tiers with additional features are coming soon.
Does flexx include agent memory?
Yes. flexx ships a graph-backed memory layer. Agents can store and recall decisions, preferences, and project context across sessions and tools — all through the same MCP endpoint.
Which platforms are supported?
flexx runs on macOS (Intel and Apple Silicon), Linux (x86_64 and ARM64), and Windows (x86_64 and ARM64). It's a single Go binary with zero dependencies.
What is the Router, and do I have to use flexx's models?
No. Router is a bring-your-own-model gateway. Add an Anthropic, OpenAI, OpenRouter, Groq, or any OpenAI-compatible key in the dashboard, set per-key budgets, and use those models inside Flexx Agent — or point your own code at the router's OpenAI- and Anthropic-compatible endpoints.
What's the difference between using Remote/Memory directly and using Flexx Agent?
Remote, Router, and Memory are independent MCP servers you can plug into Cursor, Claude, or any client today. Flexx Agent is a unified assistant that already has all three connected, so you don't have to configure each one separately.
Install the CLI
Install once. Use that machine anywhere.
Run this on the machine you want to reach. Works on macOS, Linux, and Windows — one binary, zero dependencies.
$ curl -fsSL https://cli.flexx.dev/install.sh | shirm https://cli.flexx.dev/install.ps1 | iex- 2. Connect it to your account:
flexx setup - 3. Bring it online:
flexx connect— orflexx service installfor always-on
Only need Memory? Skip all of this — create a free account and connect any MCP client. No install, no machine required.
Already have an account? Sign in · Installation guide