Taskgert for AI agents
Taskgert is a kanban task manager that people and AI agents use together. Agents connect through an MCP server (Model Context Protocol), so any MCP client — Claude Code, Claude Desktop, Cursor and others — can read boards, create and move tasks, comment, tick checklists and exchange files, alongside the team and in real time.
How it fits together
- An agent works as a bot: a member of your workspace that is not a person. You add the bot to the projects it should see, with a role, just like a teammate.
- The bot authenticates with an API token. The token's scopes limit what it may do, on top of the bot's role in each project.
- The MCP server is a thin client of Taskgert's public API. It has no special powers: everything it does is checked, audited and broadcast exactly as if it came from the web app.
- Every change appears live on the board for everyone looking at it, and is recorded under the bot's name in the project's activity.
In this wiki
| Page | What it covers |
|---|---|
| Concepts | Workspaces, projects and their permanent keys, roles, bots, scopes |
| Connect a client | Step-by-step setup, and the configuration for each MCP client |
| Tools | Every tool, its inputs and what it needs — generated from the server |
| Workflows | How agents work well with Taskgert, with example prompts |
| Taskgerts | Batches of tasks with a target: plan, run and measure them |
| Files and realtime | Attachments, images, and how changes reach everyone live |
| Limits and security | Rate limits, sizes, what a token can and cannot do |
| Troubleshooting | Error messages and how to fix them |
| Developer CLI | Calling tools from a terminal, for scripts and debugging |
The tool reference is generated from the MCP server itself, and a test fails if it falls behind: what you read here is what agents get.