Self-hosted MCP gateway

Your AI, connected to every tool your company works with.

Ask once: it reads the Jira ticket, checks the logs in Grafana, and posts the answer to Slack. One secure endpoint for Claude, Codex, ChatGPT and Gemini.

  • PII pseudonymized before the model sees it
  • Grounded in your company’s context
  • 50+ connectors, plus your own APIs
  • Every tool call audited, with pseudonymized actors
  • Service tokens stay in your gateway: AES-256-GCM, optional Cloud KMS

Free for up to 5 users.

Confluence Jira M365 Notion Slack … +43 more Figma Workspace GitLab You + AIclient HubSpot Grafana

More productivity, less copy & paste

Connect everything with everything.

One prompt across your tools You ask what is blocking the Q2 release. The question goes through the AI client to mcpgate, which reads Jira issue PBE-2674 and GitLab merge request 891. The answer comes back: PBE-2674 waits for the merge request, and Lena reviews it. You ask the AI client to message Lena in Slack. The client asks you to allow it; you allow it; the message goes through mcpgate to Lena in Slack, and mcpgate reports that it was sent. You AI client mcpgate Jira GitLab Slack What’s blocking the Q2 release? PBE-2674 MR !891 PBE-2674 waits for MR !891; Lena reviews it Ask Lena in Slack Allow: message to Lena? Allow @Lena: can you review MR !891? Sent ✓

mcpgate reads Jira and GitLab for you and writes to Slack. Your AI client asks before writing, as you set it up; for risky actions, such as deleting a whole project, mcpgate asks too.

See all 50+ connectors →

Customer data, masked before the AI sees it

Two layers: the gateway masks what its tools return, for every AI client. maisecrets masks what Claude Code, Cowork and Codex read on your laptop.

maisecrets in Claude Code: you see the value, a click copies it

From a support ticketYou seeThe AI gets
Name Laura Becker 🔒 ⟦NAME_c1⟧
E-mail laura.becker@kunde.mcpgate.de 🔒 ⟦EMAIL_c2:la•••@•••.de⟧
Password ⟦SECRET_c3⟧ 🔒 ⟦SECRET_c3⟧
What maisecrets masks, in Claude Code, Cowork and Codex
Masked dataRules (default)+ name service
Keys, tokens, passwords✓✓
E-mail, IBAN, card, phone (+49 …)✓✓
Names after a label or in a column✓✓
Names in running text–✓

The rules run on every computer, with no model to download. For names in running text, add the name service: on your computer, or once in your mcpgate.

How it works, and its limits →  ·  Set it up for your team →

Context is king, not the model

Ever gotten a confident answer that was just wrong? That’s missing context.

Confidence — how sure it sounds Actual quality — needs context the gap = overconfidence gap persists it often parks here — it doesn't feel the need to look things up Context / data the AI has → level (low → high)
An answer sounds just as sure whether it's grounded or guessing. Without your context the AI parks at high confidence, low quality — the gap is overconfidence, and it never fully closes.
shared (company) + personal context moves you right Confidence Actual quality Context / grounding →
Two layers close it: a shared, curated company map moves answers right, and personal context moves it further — quality climbs toward confidence. The gap shrinks; it doesn't vanish, and we don't pretend it does.

An answer sounds just as sure whether it's grounded in your reality or guessing. Without context, the AI parks at high confidence, low quality — that gap is overconfidence.

A shared, curated company map moves answers right; personal context moves it further — quality climbs toward confidence. The gap shrinks; it doesn't vanish, and we don't pretend it does.

What context engineering is, and why it closes the gap →

Context Map: what exists, and when did it change?

Your AI asks the Context Map for where and who, and stays current.

How the Context Map reaches your AI You start a session. mcpgate tells the AI client to search the Context Map first for where and who questions. You ask where the onboarding guide is; the question goes through the AI client to mcpgate, which answers from the map index: HR onboarding. Later a colleague updates the HR onboarding page. Your next request, for an unrelated Jira issue, goes through to mcpgate, and the Jira response comes back with the map update. You AI client mcpgate Colleague Start a session Where or who? Search the map first Where is our onboarding guide? Map index: HR onboarding Update: HR onboarding Jira: show issue SUP-214 Jira response + map update: HR onboarding

Your team updates the map by asking its AI, over MCP; every change is versioned in git. A change reaches every open session with its next tool call, without an extra request.

Why a map and not RAG →

Room for the right context

Fewer tokens per connected tool. Plus per-action audit, PII scrubbing and policy hooks raw MCP doesn’t give you. Drag to model your own stack:

How much context do your connected tools cost?
Direct MCP — curated tools resident
–
tokens, re-sent every turn
Via the gateway — resident
–
tokens — flat per service

Full calculator & method →

Speaking of the right context — only a lean set of tools stays resident. The rest — thousands of actions across every connected service — stay out of your window, discovered on demand via search when the AI needs them. That’s why each added service costs ~550 tokens here, not 11k.

Automations run through the same door

Windmill, n8n or your own agent: same rights, same rules, same audit.

An automation through mcpgate You ask for an automation: new Notion pages to Slack. mcpgate gives the AI client the Windmill tools and how to build it. The client creates the flow in Windmill through mcpgate; it runs every Monday at 8 as you. Later Windmill asks mcpgate for the new Notion pages; mcpgate reads them as you and returns them with personal data masked. Windmill asks mcpgate to post the summary, and mcpgate posts it to Slack as you. Every call is in the audit log. You AI client mcpgate Windmill Notion Slack Automate: new Notion pages to Slack Windmill tools + how to build it Create the flow: Mondays 8:00, runs as you Done: every Monday at 8 Mon 8:00: new pages? as you: read new pages the pages, PII masked Post the summary as you: post to #team

Every call is audited. Flows built in n8n, Zapier or Make can call the gateway too.

Windmill jobs through mcpgate →

Run it on your own infrastructure

One command. Self-hosted. No .env file needed.

Terminal
curl -fsSL https://mcpgate.de/install.sh | bash
  • Source-available — read what runs
  • Self-hosted
  • Your prompts are not logged

Reviewable shell script. The gateway runs on your own infrastructure.

  1. 1Run the command aboveit checks Docker and starts the stack
  2. 2Open localhost:8642the wizard signs you in
  3. 3Connect your AI clientClaude, Codex, Gemini or ChatGPT
  4. 4Set up your servicesin the admin UI, or let your AI do it over MCP

Try the demo Full quickstart →

Demo with sample data: no signup, no credit card.