> ## Documentation Index
> Fetch the complete documentation index at: https://docs.doers.sh/llms.txt
> Use this file to discover all available pages before exploring further.

# The MCP server

> Doers as native tools in Claude, ChatGPT, Cursor and VS Code.

The [Model Context Protocol](https://modelcontextprotocol.io) is how an AI assistant uses outside
tools. The Doers MCP server gives an assistant your workspace: it can read your priorities, plan
your day, create and file tasks, and tick them, in the conversation you are already having.

```text theme={null}
https://app.doers.sh/mcp
```

<Card title="Connect a client" icon="terminal" href="/mcp/connecting">
  Claude Code, Claude Desktop, claude.ai, ChatGPT, Cursor, VS Code, or any client from a JSON
  config.
</Card>

## What it exposes

<CardGroup cols={3}>
  <Card title="186 tools" icon="wrench" href="/mcp/reference/tasks">
    Every public operation but seven: tasks, projects, areas, planning, calendar, coach, sharing,
    agents and account.
  </Card>

  <Card title="5 resources" icon="file-lines" href="/mcp/reference/resources-and-prompts">
    Views to pin in a conversation: today's context, open tasks, the day's plan, projects, one task.
  </Card>

  <Card title="4 prompts" icon="message" href="/mcp/reference/resources-and-prompts">
    Ready-made workflows: plan my day, brain dump, start my round, triage.
  </Card>
</CardGroup>

The seven operations that are REST only: the five billing flows, the coach's own reply turn and the
link preview.

## MCP or REST

Both come from the same list of operations, so a tool and its REST route take the same input and
return the same output. What differs is who calls them.

|            | REST API                          | MCP server                                         |
| ---------- | --------------------------------- | -------------------------------------------------- |
| For        | scripts, n8n, cron jobs, backends | conversational assistants                          |
| Credential | a personal token                  | a personal token, or OAuth sign-in                 |
| Discovery  | you read the reference            | the assistant receives the tools and their schemas |
| Failures   | HTTP status and error code        | a message the model reads and acts on              |
| Framing    | yours to write                    | sent by the server when the client connects        |

That last line matters. When a client connects, the server sends instructions that the client puts
in the model's context: `when` is not `deadline`, a task is in a project or an area, the server
works in UTC, and the method (start with `get_context`, find identifiers with `search` rather than
asking, batch similar changes, propose a plan before committing it).

## An example

> **You**: What should I work on today? Then plan my day.

The assistant chains, without asking you for an identifier:

1. `get_context`: your counters, and your top tasks with the reasons behind each score;
2. `suggest_day_plan`: a proposal split into rounds, written nowhere;
3. it shows you three to five tasks and waits for your answer;
4. `plan_today`: saves the plan you agreed to, shown at the top of Today in the app.

## Good to know

* **Every tool acts with your account's full authority.** Neither a token nor an OAuth connection
  narrows what a client can do. Grant access only to clients you trust, and prefer clients that ask
  before running a tool marked destructive.
* **Stateless.** No session: each request carries its credential.
* **Product only.** This server and these pages cover the Doers product. The Doers website's
  content management resource is operational infrastructure: it is not public, not documented here
  and not listed in the public MCP Registry.
* **Your day.** Tools that take a `day` expect it in your time zone; without it, the server uses its
  UTC day.