Local MCP. Securely connected.

Your local tools.
Your AI, connected.

A polished desktop control center for MCP profiles, secure tunnels, validation, live status, and recovery — so connecting ChatGPT to local tools feels like a product, not a terminal ritual.

Windows release status See how it works

Local-first workflow OpenAI Secure MCP Tunnel No terminal juggling

MCP AI PlugIn Tunnel Manager

ACTIVE PROFILE

mirin-mcp-local

Connected

✓ProfileReady

✓ValidationPassed

↗TunnelConnected

✦AIActive

Runtime Healthy

Project folder/home/dev/projects/mirin

MCP targetserena start-mcp-server

Tunnel IDtunnel_6a85••••80bc6

Live consoleClear

16:08:31PASS profile_load mirin-mcp-local.yaml

16:08:31PASS runtime_api_key secured

16:08:32PASS tunnel_id configured

16:08:33INFO secure tunnel connected

16:08:34AI client session active

✓All checks passedReady to connect

✦AI ActiveLocal MCP in use

01One control centerProfiles, tunnels & runtime

02State-aware controlsActions match runtime status

03Live visibilityValidation, logs & AI activity

04Fast recoveryLess babysitting, more building

LESS FRICTION. MORE FLOW.

Local MCP should not feel
like infrastructure work.

Stop keeping setup commands, terminal windows, tunnel IDs, and connection state in your head. Put the operational layer where it belongs: behind a clear, dependable interface.

BeforeManual workflow

$ set CONTROL_PLANE_API_KEY=••••

$ tunnel-client doctor --profile...

$ serena start-mcp-server...

$ tunnel-client run...

error: connection interrupted

Commands everywhere. State nowhere.

With Tunnel ManagerManaged workflow

Mmirin-mcp-localConfiguration loadedReady

Runtime key✓ Valid

Tunnel✓ Connected

MCP server✓ Running

✓ Validated & Running

One profile. One status. One place.

BUILT FOR DAILY AI DEVELOPMENT

The operational layer
your MCP workflow was missing.

Everything needed to configure, validate, connect, observe, and recover local MCP sessions — without turning your desktop into a wall of terminals.

Profiles

Every MCP workspace, organized.

Save project folders, tunnel IDs, runtime keys, ports, and launch settings per profile. Switch contexts without rebuilding your setup every time.

Doctor & Validation

Know what is ready before you run.

Validate configuration, keys, tunnel reachability, MCP targets, and runtime dependencies with clear pass/fail states before starting a session.

Secure Tunnel

Bridge AI to local tools securely.

Launch your local MCP server through an OpenAI Secure MCP Tunnel without exposing a raw local service directly to the public internet.

Live Monitor

See the connection breathe.

Follow streaming logs and state transitions across Ready, Connected, AI Active, and Standby so you always know what the agent is doing.

Auto Recovery

Designed to recover, not just fail.

Detect interrupted tunnel or MCP processes and recover quickly, reducing the need to babysit command-line windows during long AI coding sessions.

Desktop Control

Quiet when you want it. Visible when you need it.

Run from the system tray, surface meaningful status notifications, and manage multiple active profiles from one focused control surface.

FROM LOCAL PROJECT TO AI

A secure path.
A visible state.

Tunnel Manager coordinates the pieces that usually live across scripts and terminals, while your source code and local MCP tools stay on your machine.

Select a saved profileProject, runtime key, tunnel ID, and MCP target load together.

Validate the complete chainSee exactly what passes and what needs attention before startup.

Start and observeLaunch the MCP process and tunnel, then monitor live activity from one UI.

Connection architectureLOCAL-FIRST

AI CLIENT

✦

ChatGPTMCP client session

Active

Encrypted tunnel

SECURE BRIDGE

OpenAI Secure MCP TunnelManaged by Tunnel Manager

Connected

localhost / stdio

YOUR MACHINE

</>

MCP ServerSerena / custom tools

Running

⌁

Your projectSource code + local context

DOCTOR

Validation you can actually read.

5 / 5 passed

01Profile configuration✓Passed

02Runtime API key✓Passed

03Tunnel ID✓Passed

04MCP target✓Passed

05Local runtime✓Ready

NO MORE SILENT BUTTONS

The interface reflects the real runtime state.

Controls enable and disable with the current process state. Validation remains visible. Running actions cannot be started twice. The UI tells you what the system can do right now.

✓ Ready⌁ Connected✦ AI Active☕ Standby

SECURITY BY WORKFLOW

Your local environment stays local.

Tunnel Manager is designed around the secure tunnel workflow rather than opening a raw local MCP endpoint to the internet. Sensitive runtime values stay in the desktop control layer while the tunnel provides the bridge.

Secure tunnel pathUse OpenAI Secure MCP Tunnels instead of public raw ports.

Runtime key handlingKeep platform credentials associated with the profile that needs them.

Local process visibilitySee exactly which local target is configured and running.

DESKTOP-FIRST

Built for the machine where your tools live.

The current desktop workflow centers on Windows development environments, with the product architecture evolving toward broader desktop coverage.

WindowsPrimary desktop target

Available

●

macOSCross-platform roadmap

Planned

⌃

LinuxCross-platform roadmap

Planned

GET CONNECTED WITHOUT GUESSWORK

From install to AI Active
in four visible steps.

The manager keeps setup, validation, startup, and runtime state in one flow so the first connection is understandable and repeatable.

Create a profile

Choose the local project folder and save the MCP target, tunnel ID, runtime key, and launch settings together.

Validate everything

Run Doctor before startup to verify the profile, credentials, tunnel configuration, MCP target, and local runtime.

Start the connection

Launch the MCP process and secure tunnel from one state-aware control surface.

Connect your AI client

Add the generated connector in ChatGPT, then watch the manager surface Connected, AI Active, and Standby states.

SYSTEM REQUIREMENTS

What you need on the machine.

Keep the runtime close to the code and let Tunnel Manager coordinate the connection layer around it.

Operating systemWindows 10 / 11 or Windows Server

NetworkOutbound HTTPS access to OpenAI services

MCP runtimeSerena or another compatible local MCP server

CredentialsOpenAI platform key and Secure MCP Tunnel configuration

FAQ

The questions that matter before you connect.

Clear expectations around local access, profiles, validation, and platform support.

Does Tunnel Manager expose my local MCP server directly to the internet?

No. The intended workflow uses an OpenAI Secure MCP Tunnel instead of publishing a raw local MCP port publicly.

Can I keep multiple projects configured at the same time?

Yes. Profiles are designed to keep project-specific folders, tunnel settings, keys, ports, and launch commands together so you can switch contexts quickly.

What happens if validation fails?

The failed check remains visible and start actions can stay unavailable until the required configuration is ready. This keeps the UI aligned with the real runtime state.

Is macOS or Linux available?

Windows is the current desktop target. macOS and Linux are shown as roadmap targets and should not be treated as released builds yet.

MCP AI PLUGIN TUNNEL MANAGER

Make the connection layer disappear.

Spend less time managing tunnels and terminals. Spend more time letting AI work with the tools and projects on your machine.

Windows installer — release preparingNo public installer is published yet. View release status →

Your local tools.Your AI, connected.