Connect your AI

Model Context Protocol is the standard an assistant uses to talk to external tools. There is a hosted MCP server here: nothing to install, nothing to deploy.

One URL, that’s it

Paste this address into your client. It works with Claude, with ChatGPT, and with any client that speaks Model Context Protocol over Streamable HTTP.

Server URL

https://app.projektrepublic.com/mcp

It authenticates over OAuth —your client sends you to the sign-in screen— or with a Personal Access Token as bearer, the same one you would use against the API.

Configuration per client

Copy the block for your client and paste it where it tells you. That is all.

Claude Code

claude mcp add

bash

claude mcp add --transport http projekt https://app.projektrepublic.com/mcp

Claude Desktop

claude_desktop_config.json

json

{
  "mcpServers": {
    "projekt": {
      "type": "http",
      "url": "https://app.projektrepublic.com/mcp"
    }
  }
}

Cursor

.cursor/mcp.json

json

{
  "mcpServers": {
    "projekt": {
      "url": "https://app.projektrepublic.com/mcp"
    }
  }
}

How many tools your assistant sees

The server does not register all of them at once, and that is deliberate: a model with hundreds of tools in front of it chooses worse. The core profile is exposed by default.

By default: 54 tools, the everyday ones.
Add ?tools=all to the URL to register all 309.
Or ask for just yours: ?tools=finance,crm.

Available domains

pm
okr
docs
crm
bi
automations
attachments
crossorg
calendar
finance
hr
admin

What it actually reaches

What limits a connector’s reach is NOT the MCP server: it is the ROLE of the token it connects with. Finance, payroll and HR are exposed, and an owner token reaches them. If you want your assistant to see less, give it a token with a lesser role — do not rely on the server to hide it. The sensitive block can be switched off server-side with MCP_EXPOSE_SENSITIVE, but it ships ON.

Rather host it yourself?

The same server runs locally over stdio, for desktop clients or for when the traffic must not leave your machine. Same variables, same token.