One MCP control plane. More than one transport.

MapleMCP normally reaches your machines through the native endpoint agent. When you deliberately configure it, the same account can also use an independent SSH recovery gateway that does not depend on that agent being healthy.

The normal path

Your AI→MapleMCP edge→Native agent→Your device

The native agent is the primary path because it can expose structured file, process, browser, mobile, diagnostic, administration, and fleet tools with explicit endpoint capabilities.

Independent SSH fallback

Your AI→MapleMCP edge→SSH Gateway→OpenSSH→Target sshd

If an agent is broken, stopped, updating, or repairing itself, a separately configured SSH path can still reach a registered host.

Separate transport

The SSH Gateway is not an endpoint agent and does not depend on the target MapleMCP service.

Keys stay put

Private keys stay inside the SSH Gateway. They are never returned through MCP.

Pinned host identity

OpenSSH uses BatchMode, IdentitiesOnly, StrictHostKeyChecking, and a dedicated known_hosts file.

Same owner controls

Emergency Stop, Safe Mode, Write Lock, Safety Level, approvals, audit, and session-loss confirmation still apply.

SSH is optional. Its tools appear only for accounts explicitly allowed to use a configured gateway. MapleMCP does not turn arbitrary internet hosts into SSH targets.

What the control plane does

1. Authenticate the account.

The MCP client signs in once. The edge knows which devices and connectors belong to that account.

2. Choose the connector.

Most operations target an enrolled native device. SSH tools target only hosts registered in the configured gateway.

3. Apply policy before execution.

Device capability policy governs native tools. Account controls and Safety Level also govern SSH mutations.

4. Return results and receipts.

The edge records safe operation metadata. The SSH Gateway also keeps its own receipt log with hashes rather than raw command or file contents.

Why two transports matter

A recovery path should not depend on the component you are trying to repair. MapleMCP can use the native agent for normal work and an independent SSH Gateway when you need a second way into a Linux/server target.

Use stable host IDs instead of sending credentials in prompts.
SSH reads are bounded; writes are bounded and atomic.
Standard/Guarded modes step up consequential SSH mutations.
Owner Unrestricted can use the privileged recovery path directly except for hard safety floors such as session-loss double confirmation.
Lockout-prone SSH operations are classified separately from ordinary remote commands.

Native agent versus SSH

Native agent

Best for structured cross-platform tools, desktop/browser/mobile work, device capabilities, and long-running managed execution.

SSH Gateway

Best for independent server recovery, bounded remote shell operations, and file repair when the native endpoint cannot help.

Common questions

Is MapleMCP just an SSH wrapper?

No. The native agent is the primary structured execution fabric. SSH is an optional independent recovery connector.

Does SSH expose my private key to ChatGPT?

No. The edge refers to a registered host ID. The gateway keeps credentials server-side.

Does one MCP connection expose every customer or host?

No. Device inventory is account-scoped and SSH tools are additionally configuration- and allowlist-gated.

Can SSH bypass Emergency Stop or Write Lock?

No. SSH mutation goes through the same account-level execution controls before the gateway is called.

Use the native agent first. Keep an independent recovery path when you need one.
Get MapleMCP