AI Assistants (MCP)

Paxper exposes a Model Context Protocol  (MCP) server so you can connect your own AI assistant — Claude, Codex, or any other MCP client — directly to your organization’s workspaces.

What you can do

Once connected, you can ask your assistant to work with your Paxper workspaces in plain language, for example:

• “Rename my prod-web security group to prod-web-sg.”
• “Add an S3 bucket for logs to the platform workspace.”
• “Which workspaces use t3.micro instances?”
• “What fields does an RDS instance need, and what are the valid engine versions?”
• “Connect this ALB to the web security group.”

The assistant reads your workspace diagrams, looks up valid values from AWS for the workspace’s region (AMIs, instance types, engine versions, and so on), and makes the same kind of canvas edits you’d make by hand — with your approval at each step, since MCP tool calls in most clients require confirmation before they run.

What it deliberately does not do

The MCP connector only edits the workspace diagram. It never touches your cloud infrastructure directly:

• It cannot run plan, apply, destroy, drift detection, or import. Those stay in the Paxper dashboard, where you review a real execution plan before anything changes in your cloud account.
• It cannot bypass your role. A Viewer connected via MCP still has read-only access.
• It cannot add every resource type — some are configured through other resources’ fields rather than added standalone (see Limits & Troubleshooting).

See How Edits Work for exactly what happens in Paxper when your assistant makes a change, and Security & Permissions for the authentication and authorization model.

Next steps

• Connecting — set up Claude, Codex, or any MCP client
• Tools Reference — every tool the assistant can call
• Limits & Troubleshooting — what’s out of scope and common errors