> For the complete documentation index, see [llms.txt](https://docs.katara.ai/about/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.katara.ai/about/connect-ai-clients/connect.md).

# Connect through MCP

Katara exposes a remote Model Context Protocol (MCP) server so compatible AI clients can work with authorized organization knowledge. This is a connection to Katara, not a dashboard for registering arbitrary third-party MCP servers.

## Before you start

* Use a client that supports **remote MCP over Streamable HTTP** and **browser-based OAuth sign-in**.
* Sign in with an account that belongs to the organization you want to use.
* Ask your administrator for the appropriate role. **MCP User** provides read-only MCP access; it does not grant access to the business dashboard or organization management.

## Connect

1. Open Katara's MCP connection page at `/mcp` and select the organization you are authorized to use.
2. Under **Connect your AI client**, copy the **organization server URL**. Use the exact URL shown; do not replace the organization identifier with a name or remove its query parameters.
3. In your client, add a remote MCP server named `katara` and paste that URL.
4. Start the connection and complete the browser sign-in with your Katara account.
5. Review the organization and requested permissions before authorizing.
6. Refresh the client's available tools. The server's discovered tool list is authoritative for that deployment.

For clients that accept an `mcpServers` JSON configuration, the connection page also provides **Copy JSON config**. Merge that entry with your existing servers instead of replacing your whole configuration. Other clients may use a different configuration format.

**Do not paste an API key, access token or Authorization header into this configuration.** The supported connection flow lets the client handle authentication.

## Check the connection

Start with a read operation offered by the server and a source you are allowed to access. Successful sign-in does not grant access to every collection or Brain, and a readable source does not necessarily permit writes.

For tool families and source selection, see [Tools and source selection](https://docs.katara.ai/about/connect-ai-clients/tools.md).

## Connection problems

| Symptom                                      | What to check                                                                                                                                                                                     |
| -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| No authorized organization is available      | Confirm membership with an organization administrator and sign in with the invited account.                                                                                                       |
| “MCP endpoint is not configured”             | An administrator must configure the application's `NEXT_PUBLIC_MCP_SERVER_URL` with the full endpoint ending in `/mcp`, without a trailing slash. This is a deployment setting, not a user token. |
| Client cannot finish authorization           | Confirm support for Streamable HTTP and browser OAuth; restart authorization through the client rather than copying tokens manually.                                                              |
| Connected, but a source is missing or denied | Confirm the selected organization and source permissions. MCP access and content access are separate checks.                                                                                      |
| Read succeeds but a mutation fails           | Check both the organization's MCP write eligibility and permission on the target. MCP User is read-only.                                                                                          |
| A tool described here is absent              | Refresh discovery and ask the administrator whether that capability is enabled in this deployment. Do not substitute a retired tool name.                                                         |
