Use case

Secure GitHub access for MCP-connected agents.

Direct MCP-to-GitHub access can turn a coding agent into a runtime identity with broad tool reach. z-gateway adds a policy layer between MCP clients and GitHub, and the same runtime model can be paired with Postgres database governance.

GitHub connector docs Sign up

The direct-access problem

When an MCP client talks directly to GitHub with broad credentials, the agent can attempt real actions before a central policy system sees them. Prompt guardrails and local config do not create a durable audit trail or enforce per-action allow and deny decisions.

z-gateway as the policy layer

AI agent
  -> MCP client
  -> z-gateway
  -> policy engine
  -> GitHub connector
  -> audit logs

Controls added to GitHub MCP workflows

Policy before GitHub

Requests are evaluated before GitHub installation tokens are minted or API calls are made.

Monitor mode

Observe what agents are trying to do before turning on blocking behavior.

Enforce mode

Block denied GitHub actions at the gateway for production workflows.

Paired with data governance

GitHub-specific controls remain useful on their own, but many coding agents also need to inspect schemas, run read queries, or propose database changes. z-gateway can place those Postgres actions in the same runtime session as repository work so review happens in one timeline.