# Forth MCP

> A hosted relay that puts local, NAT'd, or firewalled MCP servers behind a single secure URL, enabling any MCP-speaking AI client to call your tools from anywhere.

Forth MCP is a hosted relay service that bridges the gap between local MCP (Model Context Protocol) servers and remote AI clients like Claude and ChatGPT. It eliminates the need for port forwarding, VPNs, or exposing your machine directly to the internet by routing tool calls through a secure, centrally managed URL. A lightweight connector binary runs quietly in the background as a system service, and any MCP-compatible client can reach your local tools from anywhere.

## What It Is

Forth MCP sits in the MCP infrastructure layer: it acts as a secure relay between AI clients and MCP servers that would otherwise be unreachable due to NAT, firewalls, or local-only network configurations. The product is aimed at developers and teams who want to expose local or private MCP tool servers to agentic AI workflows without compromising network security or managing complex tunneling infrastructure.

## How It Works

Setup follows three steps:
- **Install the connector** — a single self-contained binary that registers as a Windows service, macOS launchd agent, or Linux systemd unit, installed directly from the dashboard.
- **Attach a connector token** — one token authenticates every local MCP endpoint added, whether an `npx` package or a plain HTTP address.
- **Call from anywhere** — point any MCP client at the Forth MCP URL and API token; the relay handles the rest regardless of where the machine is located.

## Key Features

- **Multi-endpoint connectors**: One connector token covers multiple local MCP endpoints, which can be added or removed independently.
- **Per-token tool permissions**: Each API token can be scoped to an allowlist of tools, so different AI clients or teammates only see what they need.
- **Usage dashboards and quotas**: Real-time visibility into calls by token and by tool, with configurable limits, threshold alerts, and an instant kill switch.
- **Live health monitoring**: Alerts when a connector instance or endpoint drops, with the ability to re-add deleted endpoints without losing the rest of the connector configuration.
- **Agentic workflow protection**: A grace buffer prevents hard cuts mid-task when a quota is reached, protecting in-progress agent sessions.

## Deployment Model

Forth MCP offers both a hosted SaaS relay and a self-hosted option. The self-hosted variant runs entirely on the customer's own network, governs both local and remote MCP servers, supports SSO, and provides unlimited audit log retention and tool calls. The hosted tiers range from a free entry plan to a Business tier with unlimited connectors and higher call volumes.

## Platform and Audience

The connector binary supports Windows, macOS, and Linux. The service is designed for developers building agentic workflows and teams managing fleets of MCP tools in production environments — not just single local demos. It works with any MCP-speaking client, with Claude and ChatGPT explicitly called out as supported.

## Features
- Hosted MCP relay
- Local MCP server exposure without port forwarding or VPN
- Lightweight connector binary (Windows, macOS, Linux)
- Multi-endpoint connectors with single token
- Per-token tool allowlists
- Usage dashboards and quotas
- Real-time health monitoring
- Kill switch
- Grace buffer for in-progress agent sessions
- Audit log retention
- Self-hosted option with SSO support
- Extra call block add-ons

## Integrations
Claude, ChatGPT, Any MCP client

## Platforms
WEB, CLI, WINDOWS, MACOS, LINUX, API

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://www.forthmcp.com
- Documentation: https://www.forthmcp.com
- EveryDev.ai: https://www.everydev.ai/tools/forth-mcp