Overview

MCP Write allows AI tools connected to SqlDBM to contribute changes to governed data models. A user can create a pending revision, create a branch, or create a new project from DDL.

What you can do

Create a revision on an existing project.
Create a new branch for isolated work.
Create a new DDL Object
Create a new project from a DDL script.

Permissions and setup

Connect SqlDBM to your AI tool using your personal SqlDBM MCP token.
Approve the MCP write scope when prompted. A read-only token cannot write.
Make sure MCP Write is enabled for your company.
The user must be a Modeler or Admin, and Writer or Owner on the target project.
In the AI tool, keep write/delete tools set to “Needs approval” if you want the user to approve each write request.

Example: SqlDBM MCP connector permissions

When you authenticate the SqlDBM MCP connector, you will be asked what level of access to grant. You can choose:

Read-only access: allows the AI tool to read SqlDBM project and model context.
Read and write access: allows the AI tool to read context and use MCP Write tools such as creating revisions, branches, and projects.

Write access is always explicit. A read-only authorization cannot create or modify anything in SqlDBM.

How governance works

MCP Write does not bypass SqlDBM governance. AI tools cannot merge, approve, or delete revisions. All writes use the caller’s existing permissions and land in the revisions list with the user’s name and timestamp.

Supported write tools

Tool	What it does
create_revision	Creates a pending revision on an existing project.
create_branch	Forks a project into a new isolated branch.
create_project	Creates a new project from a DDL script.

Notes

Only DDL is supported as the source format in this release.
MCP Write does not consume AI credits; rate limits apply per user.
Rejected writes do not create partial revisions.
A branch is returned as a separate project ID and should be used for subsequent writes.