Build with the MCP
Overview
The Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external tools and data instead of relying only on what they were trained on. Assistants that support it include Claude, ChatGPT (via Codex), and Gemini (via Antigravity). Kizen exposes an MCP server so a connected assistant can work directly with your Kizen data.
The current Kizen MCP server is read-only and exposes tools for your Knowledge Catalogs: an assistant can list the catalogs you can access, search across them, and open a specific document. Access always follows your existing Kizen permissions, so a connected assistant sees only the catalogs and content you are allowed to view.
Think of MCP as an AI-native way to reach Kizen, alongside the REST API. Where the REST API is built for your code to call, the MCP server is built for an AI assistant to call on your behalf. Unlike REST, MCP doesn't guarantee backward compatibility, since the assistant is expected to discover available tools each time it connects.
What You Can Do with Kizen's MCP Server
Once an assistant is connected, it can:
List the Knowledge Catalogs you have access to.
Run a semantic search across one or all of those catalogs.
Retrieve the full text of a specific knowledge document.
The current tools are read-only. The assistant can read your knowledge, but it can't create, edit, or delete anything in Kizen.
How the MCP Server Works
The Kizen MCP server has one endpoint per environment. Choose the host that corresponds to your Kizen environment:
POST https://app.go.kizen.com/api/mcp/readonlyPOST https://app.fmo.kizen.com/api/mcp/readonly
If your business uses a custom domain, use that domain's API host instead of app.go.kizen.com or app.fmo.kizen.com. The readonly portion of the URL selects the server's read-only configuration.
Requests use JSON-RPC 2.0 over HTTPS POST, and the assistant follows a simple pattern: It connects, discovers the available tools, then calls a tool and reads the result.
Authentication
For authentication, the MCP server supports OAuth 2.1 for connection to external AI clients such as Claude and ChatGPT. You connect once, sign in through the browser, choose a business, and approve, with no API key to manage.
For additional information, see the Authentication page.
Permissions and Data Scoping
A connected assistant acts as you. It can reach only the Knowledge Catalogs you have View permission on, and only within the business you connected to. Nothing is shared across businesses, and the assistant never gains access beyond your own.
Available Tools
For each tool's inputs, outputs, and behavior, see the MCP Tools page.
What's Next
Start by connecting a client in Connect an AI Assistant, then head to MCP Tools for the exact inputs and outputs of each tool.
Last updated
Was this helpful?