# MCP Server (/docs/mcp)

Orakle's pricing tools as a Model Context Protocol server for AI clients and agents.



The Orakle MCP server exposes pricing intelligence as Model Context Protocol
tools, so any MCP-capable AI client or headless agent can submit and read
pricing jobs directly.

Endpoint:

```
https://api.orakle.xyz/v1/mcp
```

It speaks the [Streamable HTTP transport](https://modelcontextprotocol.io/specification/2025-11-25/basic/transports)
(MCP spec `2025-11-25`) and is stateless — no session is kept between
requests. The endpoint accepts `POST`, `GET`, and `DELETE`.

## Auth modes [#auth-modes]

* **OAuth 2.1 + PKCE** — for interactive clients. The client discovers the
  authorization server, opens a browser for sign-in, and obtains a bearer
  token automatically. No manual token handling.
* **API key** — for headless agents. Send your key as
  `Authorization: Bearer sk_...`. Create keys in your
  [dashboard](/dashboard).

<Cards>
  <Card title="Connect" href="/docs/mcp/connect" description="Add the server to any MCP-capable AI app or agent." />

  <Card title="Tools" href="/docs/mcp/tools" description="get_pricing, submit_pricing_job, get_job_result — inputs, outputs, examples." />
</Cards>

## When to use MCP [#when-to-use-mcp]

MCP wraps the same job system as the [REST API](/docs/api) and
[SDKs](/docs/sdks/typescript); read access is strictly per-user. Pick by where
you work:

* **Inside an AI client or agent** → MCP. The model calls the tools directly.
* **Application code (TypeScript/Python)** → [SDK](/docs/sdks/typescript). One
  call wraps submit → stream → result.
* **Any other stack or wire-level control** → [REST API](/docs/api).