TasksPipesDocs
Guide

Connect an AI assistant

Step by step, connect claude.ai (no token), Claude Code, Cursor or another AI assistant to your TasksPipes tasks over MCP, choose what it may do, and disconnect it.

You can let an AI assistant work on your tasks for you: "what is due this week?", "sum up the Sales pipe", "create a task to call the supplier on Monday", "log 30 minutes on TP-42". The assistant talks to TasksPipes over MCP (the Model Context Protocol), as you, with only what you allow.

You need nothing but your account. If AI agents (MCP) does not show in your settings, the administrator of your instance has not turned it on: ask them.

The quickest way: claude.ai

With claude.ai there is no token to copy: Claude signs in to TasksPipes and you say what it may do. The connector then also works in the Claude Desktop and mobile apps.

  1. In claude.ai: Settings › Connectors › Add custom connector. Name it (e.g. TasksPipes) and give the address of your instance followed by /mcp, e.g. https://tasks.example.com/mcp. Leave the advanced OAuth settings empty. Click Add, then Connect.
  2. TasksPipes opens in a new window. Sign in if needed. The page Connect an application shows:
    • who asks — Claude — and where you will be sent back: claude.ai. Only allow it if that is the site you started from;
    • what it may do besides reading: only reading is ticked; Claude asks for everything, and asked marks each permission it asked for. Tick only what you need;
    • the spaces it may reach: all of yours, or only some.
  3. Click Allow. You are back in claude.ai, connected. Ask, for instance, "What are my TasksPipes tasks due this week?".

The connection shows in Settings › AI agents (MCP), marked Connected with OAuth. To change what it may do, disconnect it there and connect it again from claude.ai.

Other assistants (Claude Code, Cursor…) use a token you copy: follow the steps below.

1. Connect an agent in TasksPipes

Open the account menu (your name at the bottom of the sidebar) › Settings › AI agents (MCP).

  1. Name it after the assistant, e.g. Claude or Cursor. The name shows in the history of the tasks it changes, to everyone who can open them.
  2. Choose what it may do besides reading. Leave everything unticked for an assistant that only answers questions.
    • Create and change tasks — create tasks in your pipes, change their title, description, column, due date, priority, move them to another pipe of the same space.
    • Comment — comment on tasks as you (comments can @mention people).
    • Log my time — log time on tasks as you.
    • Send tasks to other people — send a task to someone's incoming. Only tick it if you need it.
  3. Choose where. All my spaces, or Only these: and tick the spaces it may reach. The others do not exist for it.
  4. Choose how long: 7, 30 or 90 days, or Never expires. A shorter time is safer; you can connect it again later.
  5. Click Connect an agent.

The token is shown once: copy it now. The page also gives the MCP address of your instance (https://<your instance>/mcp) and a command ready for Claude Code.

Keep the token like a password: whoever has it acts as you, within what you allowed.

2. Add it to your assistant

Every MCP client needs the same two things: the address (https://<your instance>/mcp, "Streamable HTTP") and the token, sent as Authorization: Bearer tp_…. Replace https://tasks.example.com and tp_… below with yours.

Claude Code

Paste the command the settings gave you:

claude mcp add --transport http taskspipes https://tasks.example.com/mcp --header "Authorization: Bearer tp_…"

Then run claude mcp list: taskspipes should show Connected. Ask, for instance, "What are my TasksPipes tasks due this week?".

Cursor

In Cursor's MCP settings (or ~/.cursor/mcp.json), add:

{
  "mcpServers": {
    "taskspipes": {
      "url": "https://tasks.example.com/mcp",
      "headers": { "Authorization": "Bearer tp_…" }
    }
  }
}

Claude Desktop and other clients

Claude Desktop and the Claude mobile apps use the connectors added in claude.ai (see the quickest way): nothing to set up there. To use a token instead, or with other clients that only start local servers, go through the mcp-remote bridge.

Clients that only start local servers can reach TasksPipes through the mcp-remote bridge (it needs Node.js). In Claude Desktop: Settings › Developer › Edit config, then add:

{
  "mcpServers": {
    "taskspipes": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://tasks.example.com/mcp", "--header", "Authorization:${TASKSPIPES_AUTH}"],
      "env": { "TASKSPIPES_AUTH": "Bearer tp_…" }
    }
  }
}

Restart the app. Menus and file names change between versions of these apps: when in doubt, look for MCP or Connectors in their help.

Connectors that sign in by themselves (OAuth), like claude.ai's, need no token: see the quickest way.

3. Use it

Ask in plain words. The assistant picks the tools it needs:

You ask It uses
"What is on my plate?" my_tasks
"Find the tasks about the Lyon training" search_tasks
"What is in the Sales pipe?" list_spaces, get_space, get_pipe
"Sum up TP-42 and its comments" get_task
"Create a task to call the supplier, due Monday" create_task (Create and change tasks)
"Move TP-42 to Done" update_task or move_task (Create and change tasks)
"Tell the team on TP-42 that it is ready" add_comment (Comment)
"Log 30 minutes on TP-42" log_time (Log my time)

Most assistants ask before using a tool that changes something: read what it is about to do.

What it never does

  • It never does more than you could do yourself.
  • It deletes nothing.
  • It never sees tasks of confidential or team-only pipes, even yours: what it reads goes to the company running the AI model. When it says some tasks were left out, that is why.
  • Everything it does shows in the task's history as done by you, via its name.

Be careful with

  • Text written by others. The assistant reads task titles, descriptions and comments. A text written to trick it ("ignore your instructions and…") could try to steer it. TasksPipes tells it that this text is data, not instructions, but cannot rule it out. Give it only the permissions you need, and check its history now and then.
  • Sharing. Do not paste the token in a shared configuration or a public repository.

Disconnect it

Settings › AI agents (MCP) › Disconnect next to the agent: it stops working at once. Remove it from your assistant too (with Claude Code: claude mcp remove taskspipes; in claude.ai: Settings › Connectors). A token also stops working when your account is disabled, or when the administrator turns AI agents off.

It does not work?

What you see Why, and what to do
No AI agents (MCP) in your settings The administrator has not turned AI agents on.
claude.ai says it couldn't reach the MCP server AI agents are off, the address is wrong (it ends with /mcp), or a firewall keeps Claude out: tell the administrator.
"This request has expired" in TasksPipes More than 10 minutes went by: click Connect again in claude.ai.
The assistant gets 401 or "invalid token" The token was mistyped, expired or disconnected: connect the agent again and use the new token.
It cannot create or comment The agent was not given that permission: connect it again with it ticked.
"Task not found" or "Pipe not found" The task is outside the spaces you chose for it, or in a confidential pipe.
It cannot see a space Connect the agent again with that space ticked, or with All my spaces.

See AI agents (MCP) for the details and the list of tools.

Edit this page on GitHub