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

# Remote MCP

> Connect an AI client to Qoory's stateless Streamable HTTP MCP server.

Remote MCP is in limited beta and uses the same access rules as the Developer
API. It is a hosted Streamable HTTP interface over documented REST operations,
not a separate data source.

## Endpoint

```text theme={null}
https://api.qoory.ai/mcp
```

Send the API key as a bearer token. The client should accept both
`application/json` and `text/event-stream` and use Streamable HTTP rather than
the older SSE transport.

## Runtime model

* The server is stateless and does not require a long-lived Qoory session.
* A tool call maps to one bounded REST operation.
* Tool names and input schemas follow the documented API contract.
* Scopes, credits, rate limits, usage records, and response contracts are shared
  with REST.
* Browser clients must also pass the origin policy; an allowed origin never
  replaces bearer authentication.

## Choosing tools

Start with search, resolve, or curated discovery when the entity is unknown.
Use the returned slug or opaque identifier for detail tools. Agents should not
fan out across every entity family or repeatedly widen limits to emulate a bulk
export.

## Failures

Authentication, scope, credits, rate limits, and operation availability follow
the REST error model. Keep the public request ID returned by the MCP response
when diagnosing a tool call.

Continue with [Qoory Skills](/developers/skills) to understand what an agent
instruction package adds on top of MCP.