Forth MCP | Connect any local MCP server to any AI client
Works with Claude, ChatGPT, and any MCP client

One point of contact for every
MCP server, local or remote

Forth MCP gives Claude, ChatGPT, or any MCP client a single secure address for every MCP server your team relies on, whether it's sitting behind your firewall or already public. Sign in to connect your local MCP to any AI client in minutes, or self-host the whole relay yourself for full control and security.

ForthMCP Cloud forthmcp.com

Forth operates the relay. Point a connector or endpoint at it and go.

AI clientClaude, ChatGPT, or any MCP client
Forth infrastructure
RelayTLS terminates here
token valid tool on allowlist
Your network
Connectorone or more, opens no inbound port
Local MCP server(s)

Only the Connector is a separate process, and it runs where you do, in your own network, never inside Forth's infrastructure. No direct-to-upstream option here by design: that would mean Forth's own hosted relay holds a credential to your upstream service, which isn't something we're willing to take on.

Self-hosted your domain

You operate the relay. Forth's infrastructure is never in the path.

AI clientClaude, ChatGPT, or any MCP client
Your infrastructure, the entire relay runs here
Relayyour TLS cert, offline license
token valid tool on allowlist
Connectorone or more, opens no inbound port
Local MCP server(s)
Virtual connectorno separate process, relay logic only
caller's own upstream login, no shared credential
Upstream serveroutside your infrastructure, the vendor's own server

Virtual connector is the relay reaching an already-public MCP server directly, with no Connector process and no shared credential. The relay's own "paste an access key" step is skipped entirely, redirecting the caller straight into the upstream service's real login. Each person's own token is stored and refreshed per caller, so that service's native per-person permissions apply too, on top of Forth's own tool masking.

What passes through, in transit

  • The call's actual arguments and result content, to route it. Unavoidable for anything relaying live traffic.
  • The requested tool's name, checked once against a token's allowlist or a caller's permission set.

What's never retained or reached

  • Call arguments or results. Logs record only method, tool, duration, and status.
  • Any inbound port into your network. The connector always dials out, never the reverse.

One point of contact

Local (tunneled) and already-public MCP servers, all audited from a single dashboard, not one credential per server, scattered.

Per-person masking

A caller's visible tools are the union of every permission their email is granted, checked on every call, not just at login.

Self-hosted adds

Your own TLS certificate, hot-swappable with no downtime. Offline-signed license, no phone-home. No Forth infrastructure in the path at all.

Running in three steps

No infrastructure to manage. The connector runs quietly in the background as a service.

STEP 1
Install the connector

One binary, installed as a background service on Windows, macOS, or Linux, straight from your dashboard.

STEP 2
Attach your connector token

One token authenticates every local MCP endpoint you add, npx package or plain HTTP address.

STEP 3
Call it from anywhere

Point any MCP client at your Forth MCP URL and API token. It just works, wherever your machine is.

Everything you need to run MCP in production

Built for real fleets of tools, not just a single local demo.

One-command install

A single self-contained binary. Runs as a Windows service, launchd agent, or systemd unit, published straight from your dashboard.

Multi-endpoint connectors

One connector token authenticates every local MCP endpoint you add, npx package or HTTP address alike. Add or remove them independently.

Per-token tool permissions

Scope each API token to an allowlist of tools, so different AI clients or teammates only ever see what they need.

Usage dashboards & quotas

See calls by token and by tool in real time, with configurable per-plan limits, threshold alerts, and an instant kill switch.

Live health monitoring

Know the moment a connector instance or endpoint drops. Re-add a deleted endpoint later without losing the rest of your connector.

Built for agentic workflows

A grace buffer protects an in-progress agent session. No hard cuts mid-task the instant a quota is hit.

Simple, predictable pricing

Flat monthly tiers with a hard usage cap. No metered billing surprises.

Free
$0
  • 1 connector
  • 500 calls / month
  • All tools, no filtering
  • 7-day audit log retention
  • Kill switch
Start free
Basic
$12/mo
  • 5 connectors
  • 5,000 calls / month
  • Per-token tool allowlists
  • 14-day audit log retention
  • Buy extra call blocks (5K/20K/50K)
Get started
Business
$69/mo
  • Unlimited connectors
  • 30,000 calls / month
  • Per-token tool allowlists
  • 90-day audit log retention
  • Org-wide kill switch, buy extra call blocks
Get started
Self-hosted

Want full control? You got it.

Run Forth MCP entirely on your own network, and govern every MCP server your team relies on, local or remote.

  • Unlimited tool calls
  • Governs local and remote MCP servers alike
  • Connector instances for anything behind NAT or a firewall
  • Everything runs on your own network
  • SSO support
  • Per-token tool allowlists
  • Full usage dashboards
  • Unlimited audit log retention
Local MCP
server
Connector
Online MCP
server
Your self-hosted
relay
Your AI
clients

Ready to connect your tools?

Free to start. No credit card required.

Get started free
Server connection lost, reconnecting... (reload the page if this doesn't clear)