Connecting Claude / Cursor over MCP

Wire your AI coding tool to your Taskachu workspace in about five minutes. Same identity, same permissions, same surface you use in the UI.

What MCP does

Model Context Protocol is an open standard for letting AI assistants reach into external systems with structured tool calls. Taskachu runs an MCP server at app.taskachu.com/api/mcp that exposes the same operations the UI does: list / read / create / update / move cards, manage columns and boards, attach labels, list members, and read / write Markdown documents.

Once your AI tool is connected, the agent becomes a citizen of your workspace — not a guest pasting context. It sees what you see and acts under your identity.

Generate a workspace token

Personal Access Tokens (PATs) are workspace-scoped. Each one has the same permissions as your account inside that workspace, but can only see one workspace at a time. To create one:

1. Open app.taskachu.com and pick the workspace you want to expose.
2. Go to Settings → API Tokens → New token.
3. Name it after the client (“Claude Code laptop”, “Cursor work”), copy the token immediately — it’s shown only once.

Claude Code (CLI)

The fastest path is the bundled claude mcp add command:

# A — one-liner (recommended)
claude mcp add --transport http taskachu \
  https://app.taskachu.com/api/mcp \
  --header "Authorization: Bearer $TASKACHU_PAT"

# B — manual config edit
# ~/.claude.json (or .claude.json at the project root)

For the manual approach, add this block to your .claude.json (we recommend the env-var pattern so the token never lands in the file):

{
  "mcpServers": {
    "taskachu": {
      "type": "http",
      "url": "https://app.taskachu.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ${TASKACHU_PAT}"
      }
    }
  }
}

Export the token in your shell profile (~/.zshrc, ~/.bashrc, or your secrets manager):

export TASKACHU_PAT="tskcu_xxxxxxxxxxxxxxxxxxxxxxxx"

Claude Desktop

Claude Desktop uses the same JSON config shape. Open Settings → Developer → Edit Config, paste the same MCP server block, save, then restart the app.

{
  "mcpServers": {
    "taskachu": {
      "type": "http",
      "url": "https://app.taskachu.com/api/mcp",
      "headers": {
        "Authorization": "Bearer tskcu_xxxxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Cursor

In Cursor: Settings → MCP → Add Server. Pick the HTTP transport and use the same URL + Authorization header.

{
  "mcpServers": {
    "taskachu": {
      "url": "https://app.taskachu.com/api/mcp",
      "headers": {
        "Authorization": "Bearer tskcu_xxxxxxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

Verify the connection

Ask your AI tool what it sees. A working connection responds with concrete data, not a generic answer:

• “What workspace am I connected to in Taskachu?” — should name your workspace.
• “List the boards in this workspace.” — returns real board names.
• “Show me the in-progress cards.” — pulls live card titles.

If the agent says it doesn’t see Taskachu, double-check the JSON is well-formed and the token hasn’t been revoked.

Rotate or revoke a token

Lost a laptop, suspect a leak, or just want to rotate as hygiene? Open Settings → API Tokens, hit Revoke on the line you want to kill, and create a new one. The revoked token starts returning 401 immediately — no propagation delay.