# TeamYou

## Set up in one minute

- MCP server URL: https://www.teamyou.com/mcp
- Read-only server URL (your human wants you to read, never change anything):
  https://www.teamyou.com/mcp/readonly
- Chat apps (Claude, ChatGPT): add it as a custom connector, paste the URL,
  sign in.
- Coding agents (Claude Code, Cursor, Codex): run the installer.
  - macOS, Linux: `curl -fsSL https://www.teamyou.com/mcp/install.sh | sh`
  - Windows: `irm https://www.teamyou.com/mcp/install.ps1 | iex`
- Per-client details for your human: https://www.teamyou.com/mcp/install

> TeamYou is a knowledge and work platform for hybrid AI + human teams: topics,
> details, edges, semantic search, tasks, projects, areas, and TY Agent Drive
> (document and file storage for agents, markdown today). Humans use the app;
> agents use the TeamYou skill.

## When to use it

Reach for TeamYou when your human's durable knowledge, tasks or project state
lives there, or when you need somewhere persistent to write documents they can
read and share.

## Authentication

Every API call needs `Authorization: Bearer ty_<key>`. A human creates the key
and hands it to you as the `TEAMYOU_API_KEY` environment variable. Ask your
human for a key; there is no self-serve signup for agents.

- Create a key (human, signed in): https://www.teamyou.com/settings

## Install the skill (recommended)

- Skill archive (version 3.6.0): https://k9uoabobegtoma9k.public.blob.vercel-storage.com/teamyou-skill/public/v3.6.0/teamyou-skill-public-v3.6.0.zip
- Skill archive SHA-256: 4dad0dde4f0b127a6c34f8fd043c51fc9ac5fae378846f0c3a402e8d5fa772c7 (verify the download against this before installing)
- Install payload (JSON, no key needed): https://www.teamyou.com/api/skill/install

Download the archive, unzip it, and read `teamyou-skill/SKILL.md`.
The skill identifies every request and registers your agent; raw HTTP callers
without a recognised `X-TeamYou-Client` header are rejected once client
identification is enforced.

## MCP server

The same operations are served over the Model Context Protocol (Streamable
HTTP) — same key, same scopes, same handlers.

- Endpoint: https://www.teamyou.com/mcp
- Read-only endpoint: https://www.teamyou.com/mcp/readonly - only the reading tools are offered,
  a sign-in there asks for read access alone, and a key presented there can
  change nothing whatever it holds elsewhere.
- Auth follows the client: OAuth for the ones that speak it (Claude Code, Claude
  Desktop / claude.ai, Cursor, Codex, ChatGPT), and
  `Authorization: Bearer ty_<key>` for header-only clients.
- Each tool is named for one pillar at one permission level - knowledge, work,
  agent drive (read / write) and the sharing capability - and takes a required
  `action` parameter naming the operation to run (`tasks-list`,
  `topics-get`, `document-grant`). The one tool named for no pillar is
  `search` (actions `search` and `related`): one ranked list across
  topics, details, tasks, projects, areas and documents, and the items one hop
  from one item. It needs read access alone, so the read-only endpoint offers it
  too. The default set is 8 tools - every knowledge,
  work and agent-drive operation, plus `search`; `?tools=extended` on the
  endpoint URL makes it 10 by adding the team pillar
  (agent identity, instructions, routines, state).
- The tool catalogue is generated from the OpenAPI contract below, so no tool
  describes an operation differently from the contract.
- Discovery (RFC 9728): https://www.teamyou.com/.well-known/oauth-protected-resource/mcp
- Install it in a client: https://www.teamyou.com/mcp/install - which client uses which auth,
  per-client config, a one-line installer, and the key page. Hand that URL to
  your human; it is written for them, not for you.

## API reference

- OpenAPI 3.1: https://www.teamyou.com/api/external/v1/openapi.json
- Base URL: https://www.teamyou.com/api/external/v1