← All guides
Tutorials

Connect an AI agent over MCP

Kstonebase speaks the Model Context Protocol (MCP), so your AI agents can read and update your specifications directly. This guide connects an agent in the two supported ways.

Open developer settings

Go to Settings → Developer. This is where you manage MCP access — a ready-to-paste agent config, personal access tokens, and remote MCP clients.

The Developer settings page with a Connect an MCP agent config block.The Developer settings page with a Connect an MCP agent config block.

Add Kstonebase as a remote connector

For Claude.ai (or Cursor, ChatGPT, etc.), paste the remote MCP client URL into the tool and complete the consent flow on Kstonebase. New workspaces and products appear automatically if you pick “All workspaces” during consent.

The Personal access tokens and Remote MCP clients sections of the Developer page.The Personal access tokens and Remote MCP clients sections of the Developer page.

Or create a personal access token

For a self-hosted MCP server or other clients, click Generate token, name it, set an expiry, and optionally scope it to specific products. Copy the value once — it authenticates the MCP server with the Kstonebase API.

The Generate token dialog with a name, expiration options, and product scope.The Generate token dialog with a name, expiration options, and product scope.

Point the agent at a product

Drop the token into the MCP server config so the agent works in the product you choose. It can now list, read, and — with write scope — update specifications.

Verify the connection

Ask your agent to list your specifications. A successful response confirms it's reading from your live knowledge base.