# TablePort for agents

This reference is for LLMs, automation agents, and integration clients. For
restaurant staff performing tasks on screen, use [Restaurant guides](/).

## Start with capability discovery

1. Connect using the operator's authorised TablePort account.
2. Discover tools with MCP `tools/list`. Use the returned names and schemas for
   this connection; they determine what is available.
3. Resolve the business and location from authorised context and returned
   records. Never invent identifiers or reuse another tenant's identifiers.
4. Use [supported capabilities](/agents/capabilities) to understand the external
   connector boundary.
5. Follow [operational workflows](/agents/workflows) and
   [recovery rules](/agents/recovery).

Feature availability depends on the connected environment, release, account, and
permissions. A restaurant UI action, internal assistant capability, or skill
description does not make the same action available to an external client.

## Authorisation

The restaurant connector uses OAuth. Business read and write scopes are named
`mcp:business:read` and `mcp:business:write`. A scope is a prerequisite, not a
promise that every possible read or write is exposed. TablePort also enforces
the user's access, feature entitlements, and operation-specific rules.

Do not request staff PINs, passwords, provider secrets, or copied bearer tokens
in conversation. Follow the connection's authorisation flow. Platform
administration is outside this restaurant integration guide.

## Machine-readable entry points

- [Agent topic index](/_docs/agents/llms.txt)
- [Agent reference as Markdown](/_docs/agents/llms-full.txt)
- [Versioned business tool schemas](/_docs/agents/business-tool-contracts.json)
- [Schema and result conventions](/agents/contracts)
- [All published topic metadata](/_docs/manifest.json)

Each reference page has a Markdown version under `/_docs/`. A documentation
export contains instructions, not restaurant account records or credentials.