Built in, not bolted on

AI that proposes. You still decide.

TriggerFlo has an assistant you can ask about your own work, a microphone that turns a ramble into tasks, and an MCP server so your own AI agent can work the board directly. None of it writes to your board without you saying so — except the one that is supposed to, and we are explicit about which.

Download free Connect an agent

Flo, the in-app assistant

A second opinion on your own board

Flo sits inside the app and can see the work you have already captured, so you can ask it things a chatbot in another tab could never answer.

Ask about your own work

What is overdue, where Tuesday went, which project ate the week. Flo can read your projects, columns, views, tasks, subtasks, and comments to answer.

Break down a vague task

Hand it something you have been avoiding because it is too big, and get subtasks small enough to start.

Pick the model

Where the app offers a picker, the model you choose is tried first. Requests run on Cloudflare Workers AI and fall back to Anthropic when it is unavailable.

It cannot change your board on its own

Flo can propose new tasks, but it cannot create, change, or delete anything without your explicit confirmation. Every write shows you what it wants to do first. Nothing lands on your board because a model was confident.

You are asked to consent before any of your content is sent to an AI provider, and you can withdraw it any time in Settings. The privacy policy names every model we use and what is sent to it.

Voice capture

Say it once, get a list

Five things in your head and no interest in filling out five forms. Tap the microphone, say all of it in one go, and TriggerFlo splits the transcript into separate tasks with titles and notes. You review the list before anything is created.

Dates stay as you said them

Say "next Friday" and that phrase is kept as a hint on the task rather than resolved to a date. The parser is told not to guess. A wrong due date that looks precise is worse than a vague one you can fix in two seconds.

Where it works

macOS, Windows, iOS, and Android. Voice capture needs platform speech recognition, so the microphone is hidden where it cannot run — Linux has no support for it, and on the web it is Chrome and Edge only.

MCP server

Point your own agent at your board

TriggerFlo speaks the Model Context Protocol, so Gemini, Raycast, Claude Code, or anything else that speaks MCP over HTTP can read and update your work directly. No copy-pasting between windows. Connect with an API key, or just sign in.

Connect it

Create an API key in the app under Settings → MCP & API, then point your client at the server. Every client needs the same three things — HTTP transport, the URL, and the key as an X-API-Key header.

Raycast

Tested

Run Manage MCP Servers → Install MCP Server, set Transport to HTTP, paste the URL below, and add one HTTP header. Leave the OAuth fields empty. Then type @TriggerFlo in any Raycast AI chat and ask for what you want.

Transport      HTTP
URL            https://triggerflo.app/mcp
HTTP Headers   X-API-Key   tflo_your_key_here

Claude Code

claude mcp add --transport http triggerflo https://triggerflo.app/mcp \
  --header "X-API-Key: tflo_your_key_here"

Clients configured by file

{
  "mcpServers": {
    "triggerflo": {
      "type": "http",
      "url": "https://triggerflo.app/mcp",
      "headers": { "X-API-Key": "tflo_your_key_here" }
    }
  }
}

What the agent can do

Projects and boards
get_projects, get_project, get_project_columns, get_views
Tasks
get_tasks, get_task, create_task, update_task
Subtasks
get_subtasks, create_subtask, update_subtask
Comments
get_comments, add_comment
Attachments
download_attachment

The key carries your own permissions — an agent reaches exactly the projects you can, and nothing else.

Or sign in, instead of pasting a key

Clients that follow the MCP authorization spec — Gemini and Claude’s custom connectors among them — don’t need a key at all. Give them the server URL and leave every other field blank. They discover the rest themselves, and you approve the connection in a browser the way you would any other app.

1. Paste the URL

https://triggerflo.app/mcp — nothing else. If the client offers optional client ID and secret fields, leave them empty; it registers itself.

2. Sign in and allow

A TriggerFlo page opens listing exactly what the app is asking for. Sign in with your password, Google or Apple, then choose Allow or Deny.

3. Done

The app appears under Settings → MCP & API → Connected apps, where you can disconnect it later.

The practical difference is what happens when something goes wrong. An API key is your whole account — anything you can do through the app, it can do. A connection made this way reaches only your projects and tasks, never account settings or billing, and can be cut off on its own without disturbing anything else you have connected.

This one does write without asking

Unlike Flo inside the app, an MCP client can create and change tasks with no confirmation step. That is the point of connecting one — but it means the access is worth guarding, whichever kind you used. Revoke an API key under Settings → MCP & API, or disconnect a signed-in app under Settings → MCP & API → Connected apps. Disconnecting stops further access; a session already underway ends within the hour.

Full MCP setup guide REST API reference