# MCP

Texa as an MCP server for your team's AI tools, and your MCP server as a source of actions.


Texa works with the Model Context Protocol in both directions.

## Your AI tools ask Texa

Texa is an MCP server. An assistant your team already uses can ask about your customers,
and every answer is backed by real conversations. It connects over Streamable HTTP:

- **URL:** `https://api.texadesk.com/mcp`
- **Header:** `Authorization: Bearer texa_live_…`

| Tool | What it does |
|---|---|
| `ask_customers` | Answers a question such as "what are customers complaining about this week?" from your conversations, citing the conversation and the exact words for every claim. When the conversations do not say, it answers that it cannot answer from the data. Needs a live key |
| `list_conversations` | Recent conversations, by state or customer |
| `get_transcript` | What was said in one conversation |
| `support_report` | The support report for the last few days |
| `usage` | What has been counted this month |

Every tool only reads. Tool calls count as requests, and each question as an Ask AI
question, against your credits. They appear in your request log.

## Your MCP server runs Texa's actions

If your systems already speak MCP, add a connector of type **MCP server** in Texa Desk
(**Texa AI → Actions → Connectors**), with your server's `https` address and, if it needs
one, a bearer token. **Import tools** lists your tools as draft actions. Review each one
(its outputs, which fields are private, whether the customer must be verified, and for
tools that change something, that the customer must say yes), then save it. Texa AI then
asks for those actions like any other: gated, recorded, and told honestly when they fail.
A tool's `structuredContent`, or its JSON text, is the action's result, and `isError` is a
failure.