Documentation

Everything you need to deploy, configure, and operate mcpgate.

Getting Started

mcpgate is a self-hosted MCP gateway that connects AI clients (Claude, ChatGPT, Codex, Gemini) to your company tools through a single endpoint.

  • Quick Start — Deploy mcpgate with Docker in under 5 minutes. No .env file needed.
  • Services — Jira, Slack, GitLab, GitHub, Google Workspace, Microsoft 365, Notion, Confluence, Figma and the rest of your daily stack.
  • Hooks — Deterministic policy enforcement: Markdown→ADF, write-safety, PII handling. For style and template instructions, see Skills.
  • Configuration — Environment variables, YAML config, OIDC setup, and branding.

How it works

Every AI tool call flows through mcpgate's hook pipeline:

  1. AI sends a tool call (e.g. jira_write_actions → create_issue)
  2. Pre-hooks run: validate, block, transform, enrich
  3. Action executes against the service API
  4. Post-hooks run: cap responses, add hints, notify

Hooks are configured in YAML — no code required. Enable, disable, or reorder them and hot-reload without restart.

Architecture

mcpgate architecture — AI clients connect via MCP, requests flow through authentication, pre-hooks, action executor, and post-hooks before reaching company tools via OAuth2

Authentication

mcpgate supports multiple sign-in methods:

MethodUse case
Broker loginGoogle/Microsoft sign-in, zero config (default for new instances)
OIDC SSOYour own identity provider (Google, Microsoft, Okta, Keycloak, Auth0)
Magic LinksEmail-based login for external collaborators

Login and service credentials are independent. Switching your SSO provider doesn't affect connected services.

Pricing

Free for up to 5 users. View pricing plans for Team and Enterprise.