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

# MCP Server

> Connect Cursor, Claude, and other MCP clients to your Voltai knowledge agent.

Voltai provides a hosted [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) server. It lets an AI client ask questions against the same knowledge, tools, citations, permissions, and conversation history as the [Chat Query API](/api-reference/knowledge-agent/chat-query).

The server exposes one tool:

* **`ask_voltai`** — asks the Voltai knowledge agent a question and returns the answer, cited sources, and a `chat_id`.

## Connection details

| Setting | Value |
| - | - |
| Transport | Streamable HTTP |
| URL | `https://api-prod.voltai.ai/mcp/` |
| Authentication | `X-API-KEY: <key>`, or `Authorization: Bearer <key>` for clients that only offer a token field |
| Caller identity | `X-Voltai-User-ID: <stable-user-or-client-id>` |

Use the API key issued for your Voltai organization. Keys with no scope restrictions can use MCP. A scoped key needs both the **MCP** and **Chat API** scopes, because `ask_voltai` runs a chat query on your behalf.

Set `X-Voltai-User-ID` to a stable, non-secret identifier that is unique to the
person or client using this MCP connection. It keeps floating-seat accounting
and concurrent-request limits separate when multiple MCP users share one API
key. Do not put an email address or other personal information in this header.

## Cursor

Add Voltai to your project-level `.cursor/mcp.json` or your user-level MCP configuration:

```json theme={"dark"}
{
  "mcpServers": {
    "voltai": {
      "url": "https://api-prod.voltai.ai/mcp/",
      "headers": {
        "X-API-KEY": "${env:VOLTAI_API_KEY}",
        "X-Voltai-User-ID": "${env:VOLTAI_MCP_USER_ID}"
      }
    }
  }
}
```

Set `VOLTAI_API_KEY` and a unique, stable `VOLTAI_MCP_USER_ID` in the
environment that launches Cursor, then enable the server in **Cursor Settings
→ Tools & MCP**.

## Claude Code

```bash theme={"dark"}
claude mcp add --transport http voltai \
  https://api-prod.voltai.ai/mcp/ \
  --header "X-API-KEY: $VOLTAI_API_KEY" \
  --header "X-Voltai-User-ID: $VOLTAI_MCP_USER_ID"
```

## Asking questions

The client normally chooses when to call `ask_voltai`. You can also ask explicitly:

> Use Voltai to find the maximum operating temperature for this component.

The tool accepts:

* **`query`** (required) — the question to answer.
* **`chat_id`** (optional) — a previous result's ID to continue that conversation.
* **`reasoning_level`** (optional) — `low`, `medium` (default), or `high`.
* **`topic_names`** (optional) — knowledge topics to search.

Its result contains `final_answer`, `sources`, `chat_id`, `chat_url`, and `conversation_id`.

## Conversation and usage behavior

Pass a returned `chat_id` into a later call to preserve context. Calls use the same organization permissions, license limits, throttling, and source access as `POST /chat/query/`.

The MCP server runs the Voltai knowledge agent behind the tool call. This provides the full agent behavior with a small integration surface; it does not expose Voltai's internal tools individually.

<Note>
  The **MCP** option in this documentation site's copy menu gives an assistant access to documentation content. The hosted URL above connects an assistant to your organization's live Voltai knowledge agent.
</Note>