# Midplane

> Midplane is a thin access layer in front of an existing Postgres database that makes it safe for AI agents: read-only by default, writes opt-in per table, every query parsed and logged. It speaks MCP (Model Context Protocol), so agents like Claude Code, Cursor, and custom LLM apps connect to it instead of the raw database. Open core and self-hostable.

## Docs

- [What is Midplane?](https://midplane.ai/docs): overview of the safety layer between AI coding agents and Postgres
- [Quickstart](https://midplane.ai/docs/quickstart): point Midplane at a Postgres, connect an agent over MCP
- [How it works](https://midplane.ai/docs/how-it-works): AST parsing, policy decision, audit trail
- [Table access policies](https://midplane.ai/docs/concepts/table-access): per-table deny / read / read_write
- [Masking](https://midplane.ai/docs/concepts/masking): column-level masking of PII before it reaches the agent
- [Audit trail](https://midplane.ai/docs/concepts/audit-trail): every attempt logged, allowed or denied
- [Threat model](https://midplane.ai/docs/security/threat-model): what Midplane defends against
- [MCP tools reference](https://midplane.ai/docs/reference/mcp-tools): the tools exposed to agents
- [Self-hosting](https://midplane.ai/docs/self-host/deploy): run the open-core engine yourself
- [Cloud vs self-host](https://midplane.ai/docs/cloud-vs-self-host): choosing between managed and self-hosted
- [Full docs index](https://midplane.ai/docs/llms.txt): machine-readable list of all documentation pages

## Product

- [Landing page](https://midplane.ai): what Midplane is and why
- [Live demo](https://midplane.ai/demo): interactive policy playground — no signup, no real database
- [GitHub](https://github.com/midplaneai/midplane): open-core engine source

## Optional

- [Privacy policy](https://midplane.ai/privacy)
- [Terms of service](https://midplane.ai/terms)