Asking an AI to organize your week doesn't get you very far when it's working from a block of text pasted into a conversation. It produces a tidy list, you copy it back by hand, and you've lost the time you were trying to save.
Flots is opening its MCP server. Claude, and any other compatible assistant, plugs into your account and works on your real tasks, your real projects, your real notes.
What an MCP-connected assistant on your real data changes
MCP is an open standard that lets an assistant reach an application. A plain chat can give you just as accurate an answer. What changes is what happens next.
| Without a connection | With Flots connected |
|---|---|
| You describe your tasks in the conversation | The assistant reads them straight from your account |
| The answer comes as text, to be copied over | Tasks are created in Flots, dated and filed |
| The assistant has no idea what changed since yesterday | It sees the real state of your projects the moment you ask |
| One question per project, by hand | A single question that cuts across every project |
Where it's useful
The gain shows up as soon as describing what you want is faster than clicking from screen to screen. Especially when the answer spans several projects, or when one sentence has to turn into a dozen tasks.
A few requests that work well:
- "What's overdue across all my projects this week?"
- "Turn my meeting notes into tasks, one per decision, due Friday."
- "Give me the state of the Website project: progress, blockers, who's doing what."
- "Tidy the loose documents in my Client folder into a coherent structure."
That's only a sample. The assistant reaches tasks, projects and their collaborators, notes, the document tree, milestones, categories and project reports.
What protects you
Opening your organization up to an AI raises a real question of trust. Three safeguards each answer a different risk.
You choose the permissions at connection time. An assistant granted read access only can't write anything, whatever it's asked to do afterwards. You can also leave notes out entirely.
Actions with consequences never go through silently. Deleting an item, or writing something that lands on a colleague, requires the assistant to show you the exact change and get your approval before applying it.
The text of your notes stays closed by default. It's the most cautious setting on the screen, and that's deliberate: text written by someone else can contain instructions aimed at the assistant reading it. Three levels exist, from "never" to "all notes".
Note reading is off out of the box
A connected assistant sees the titles of your notes, never their contents, until you allow it yourself.
Connecting Claude to Flots in four steps
- Open your assistant's connector settings. In Claude: Settings, then Connectors.
- Choose to add a custom connector, and paste the Flots server address. It's shown in Settings, under Connected applications (MCP), with a button to copy it.
- Sign in to Flots in the window that opens, then choose what the assistant will have access to.
- Come back to this screen whenever you like: every connected application appears there with its last use, and can be cut off in one click.
The same screen hosts documentation generated from the server itself. It lists every available action, what each one reads, writes or deletes, and flags the ones that are irreversible. Because it comes from the server, it always matches what's actually deployed.
If what you're really wondering is what an AI is good for in a working day, we sorted the uses that hold up from the rest in AI assistants and productivity: what actually helps. And for the assistant built into Flots, the one that answers your questions about the app, that's Cap.
Plug your AI assistant into your real tasks.
