Connect an AI agent

dotby speaks MCP — the Model Context Protocol. Connect a client once and your AI assistant can work inside your tracker instead of guessing about it.

Not “summarize this text I pasted”. Actually read the board, create the task, attach the screenshot, post the initiative update.

What an agent can do

Roughly 50 tools, across everything the product does:

Area	Examples
Tasks	Search, read, create, update, move, archive, comment
Projects	List, read, create, update
Sprints	List, create, update, complete
Initiatives	List, read, roll up progress, post the weekly update
Pages	Search, read, create, update, organize into folders
Files	Attach a file, and read an image someone else attached
People	Resolve a member, list the roster
Time	Read entries, see the running timer

Two of those are worth calling out, because most connectors do not have them.

Files, not just text. An agent can attach a screenshot, PDF, or document to a task — and it can look at an image someone else attached. “What is wrong in this screenshot?” gets an answer, not a download link.

Strategy, not just tasks. Over MCP an agent can read your initiatives, roll up progress across every project in one, and post the health update itself.

Before you start

• MCP access is a Pro feature. A Free workspace answers with upgrade_required.
• An agent can only ever do what you can do. It signs in as you and inherits your role, your projects, and your permissions. There is no super-token.

Connect

The server URL is the same for everyone:

https://api.dotby.app/mcp

Authentication is OAuth — you approve the connection in the browser, and no key is pasted into a config file.

Tip

You do not have to hunt through this page. In the app, go to Settings → MCP: it lists 23 clients, and picking yours shows the exact snippet with your server URL already filled in, one click to copy.

Popular clients

Claude Code

claude mcp add --transport http dotby https://api.dotby.app/mcp

Then authorize:

claude mcp login dotby

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "dotby": {
      "url": "https://api.dotby.app/mcp"
    }
  }
}

Restart Cursor, then approve the OAuth prompt.

Claude Desktop

Settings → Connectors → Add custom connector, paste the server URL, and approve the browser prompt.

ChatGPT

Enable developer mode, add a connector, and paste the server URL.

Every supported client

Settings → MCP has copy-paste instructions for all of these:

CLIs — Claude Code · Codex CLI · Gemini CLI · GitHub Copilot CLI · Droid (Factory) · OpenCode · Goose · Amazon Q Developer

Editors and IDEs — Cursor · VS Code · Windsurf · Zed · JetBrains AI Assistant · Kiro · Continue · Cline · Roo Code

Desktop and web — Claude Desktop · Claude (claude.ai) · ChatGPT · Warp · LM Studio

…plus a generic entry for any other MCP client, which gives you the raw URL and transport.

Verify it works

Ask your agent something only your workspace can answer:

What are my open tasks in the WEB project, sorted by priority?

If it comes back with real task numbers, you are connected. If it asks you to paste something, it is not.

What agents can and cannot see

Sees	Exactly what your account sees — your workspaces, your projects
Does not see	Private projects you were never added to
Cannot do	Anything your role forbids. A guest’s agent is a guest.
Modules	A disabled module answers with a clear error, not silence

Every write an agent makes is attributed. It is not an anonymous edit.

Multiple workspaces

If you belong to several workspaces, tell the agent which one — by slug or by name — or let it ask. Nothing crosses between workspaces.

Next

• API keys and tokens — for scripts and CI, where OAuth is not practical.
• Build your first integration — the REST API by hand.
• Use the CLI — the same power in your terminal.