Midplane
Midplane provides an MCP gateway that sits between AI assistants and Postgres databases, intercepting every SQL statement to block dangerous queries before execution. The gateway enforces configurable policies, masks sensitive fields, and requires human approval for multi-row writes, while running inside the customer's network. It works with Claude, ChatGPT, Cursor, and any MCP client, and is MIT-licensed with source available on GitHub.
- Developer Tools
- Cybersecurity
- Artificial Intelligence
- Enterprise Software
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
AI assistants connected to production databases can generate SQL that is syntactically valid but semantically destructive, such as an UPDATE without a WHERE clause that modifies tens of thousands of rows. Without a safety layer, these high-confidence errors reach the database and cause irreparable data loss or unintended mass changes before a human can intervene.
Solution
Midplane provides a gateway that sits between AI assistants and Postgres databases, intercepting every statement and checking it against a configurable policy before execution. The gateway blocks mass updates, deletes without filters, dropped tables, schema changes, and hidden writes concealed inside read-like statements. Sensitive fields are masked inside the query itself, so an AI cannot reconstruct emails or phone numbers by asking repeated questions. Any write affecting multiple rows waits for explicit human approval, with the exact SQL and row count presented for review. The gateway runs inside the customer's network, while policy management and audit logs are accessible through a cloud dashboard or self-hosted interface. Agents reach the gateway through Midplane's relay, which passes encrypted traffic without reading it, or at a URL of the customer's own.
Target Audience
Primary users are engineering teams and platform developers building AI-powered features on Postgres databases, as well as organizations deploying Claude, ChatGPT, Cursor, or other MCP-based assistants into production environments where data integrity is critical.
Features
- Pre-query policy enforcement that blocks UPDATE/DELETE statements lacking WHERE clauses or affecting every row in a table
- Statement-level analysis that catches hidden writes, such as a DELETE embedded within a SELECT or CTE
- Query-time masking of sensitive fields, including hashed emails, shortened phone numbers, and generalized dates
- Table access denied by default, with tables opened individually for read or write operations
- Human-in-the-loop approval workflow that holds multi-row writes until an authorized user reviews the exact statement and projected row count
- Containment feature that automatically revokes access to secret tables when an agent reads untrusted external text
- Full audit logging of every statement on the gateway's local disk, with the gateway being MIT-licensed and open source
- Deployment options including local computer, Docker behind the relay, or a customer-owned URL with no inbound ports required