Layers

AI and agents

Ask, and Layers answers from your tasks

The Layers assistant finds the tasks in your workspace on its own and answers with links to them. Claude, Cursor and other agents connect over MCP and work with tasks, pages and forms.

Layers for teams: the first 5 seats are free, then $4 per seat a month

Buy seats · $4/seat

This text was prepared with the help of AI.

No card required · free for up to 5 people

AI assistant: the question “What’s on me today?” and an answer linking the workspace tasks

Documents in ChatGPT and Claude

Connect Layers Documents in one minute

Use one remote MCP address. OAuth opens in Layers, and the agent gets only the permissions of the account that signs in.

  1. 1

    Add the connector

    Open connectors in ChatGPT or Claude and paste the remote MCP address. Choose OAuth when the client asks how to sign in.
    Remote MCP address
    https://app.layers.md/mcp?profile=documents
  2. 2

    Sign in to Layers

    Approve the connection in Layers. Passwords and access tokens stay out of the conversation.
  3. 3

    Ask for the finished document

    Create from text or a template, fill known fields, open the result for editing, or send the exact document for signature.

Try it in your own words

Three document jobs, one conversation

The agent returns both the working document and a Layers action link for the next person.

“Make a lease agreement”

You: Make a lease agreement for this apartment.

Agent: I’ll ask for the missing parties, dates, address and rent, fill the template, then return the edit link and the link you can send.

“Compute final pay”

You: Compute final pay from these figures and make the document.

Agent: I’ll use only the figures you provide, create the filled document in your Drive and return it ready to review.

“Send for signature”

You: Send this document to the named recipient for signature.

Agent: After you confirm the exact person, I’ll send once and track Opened, Agreed and Signed for you.

The problem

How AI works for most teams

AI on the side

  • ✗The chatbot doesn’t know your tasks or docs
  • ✗Answers get copied into the tracker by hand
  • ✗An agent gets full access — or none at all
  • ✗No one knows who spends how much on AI

01 · Ask

An assistant that knows your tasks

An AI chat built into Layers. It searches your workspace tasks on its own and answers with links to them.

Answers

What’s on me today — in one question

The assistant finds your tasks for today and shows the priority and status of each. Every task name links to the task.

New assistant chat: “Hi Summary!” and the prompts What is on me today, Overdue tasks, Sprint summary, Who is doing what

Suggestions

It suggests where to start

A new chat opens with four suggestions: What is on me today, Overdue tasks, Sprint summary and Who is doing what.

02 · Connect

The agent works in your workspace

The Layers MCP server gives an agent tasks, pages, forms and projects. What the agent does shows up right away — as if you did it yourself.

Tasks

From plan to tasks, no manual entry

The agent breaks a plan into tasks and sets assignees and due dates. The tasks land in the list with their own keys, just like ones created by hand.

Pages

Notes from the agent go straight to Pages

The agent creates a page in the right folder and adds sections with headings and lists. “Sprint 14 retro notes” sits under “Meeting notes”.

Forms

A form from the agent, questions included

The agent builds a form: a start screen and four questions as steps.

Beta tester survey form: start screen and a strip of four question steps
Share and permissions: public link access, Unpublish, Open form

Publishing

And publishes it

Public access by link is on; one button unpublishes it.

Comments

The agent comments on a task — as you

The agent’s comment appears in the task’s Activity like your own: the author is the token owner.

Task activity: a comment summarising the feedback, posted as Emma Clarke

03 · Grant access

An agent gets exactly the access you give it

The agent signs in with a token. You set its access and expiry, and revoke it in one click.

Access

Access and expiry, set when you create the token

With Read and write access, the token creates and changes tasks, pages and comments — everything you can do yourself. Expiration: 7, 30 or 90 days, 1 year, or never.

Access tokens: Claude (MCP) read only, Cursor read and write, expiry dates, Active status and the Revoke token tooltip

Expiry and revoking

See who has what access

Every token has its own access, expiry date and status. To revoke one, use the button in its row.

04 · Extend

Your own MCP servers, in the assistant

Connect a wiki, a CRM or your own service over MCP, and the Layers assistant will work with it.

A connector is a name and an address

A name, the server address — and a token if the server requires authorization. Settings from another client can be pasted into the JSON tab.

05 · Control

The owner decides what’s allowed

AI features and everyone’s monthly limit live in the workspace settings.

Turn AI features on or off for everyone

Assistant, editor, meetings and podcasts — each feature has its own switch. A feature switched off disappears for everyone in the workspace.

A limit per person — or AI off

One monthly limit applies to everyone without a limit of their own. A person’s menu has Set a limit, Same as everyone and Turn AI off.

Cloud or your own server

Free for up to 5 people. Self-host whenever you want

In the cloud — no card, no time limit. On your own server you get the same MCP server, and your admin connects the model for the assistant.

One workspace

AI is just the start

All products →

Questions

Frequently asked questions

Didn’t find an answer? Write to contacts@layers.md.

Do I need an AI key to connect Claude or Cursor?

No. The Layers MCP server works on its own: the agent brings its own model, and Layers gives it tools and data under your token.

How do I connect an agent?

Create a token in Settings → Access Tokens, then add the Layers MCP server address and that token to your MCP client.

What can an agent do?

More than 50 tools: tasks (including in bulk), pages, forms, projects, sprints, folders, tags, comments, search and trash. With a read-only token it can only look.

Can an agent see other people’s documents?

No. The agent works with the token owner’s permissions and sees exactly what they see.

Which models can the assistant use?

On your own server, the admin connects OpenAI, OpenRouter, Gonka or any OpenAI-compatible server — including a local model via Ollama.

How do I cap spending?

A monthly limit per person — shared or individual. AI can be turned off for a specific person, or a feature turned off for the whole workspace.

Your agents work right inside Layers

Start free in the cloud or deploy Layers on your own server.

Sprint 14 in Projects: Backlog tasks with assignees and due dates