> For the complete documentation index, see [llms.txt](https://docs.rootcause.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rootcause.ai/api-and-integrations/api-access/mcp-integration.md).

# MCP Integration

Connect Claude Code, Claude Desktop, Claude on the web, or Cursor to RootCause over the **Model Context Protocol (MCP)**. The assistant works with the same objects you see in the platform, under the same permissions: explore workspaces, query data, import sources, run causal discovery, train and simulate Digital Twins, write reports.

## What to expect

* **Read-only by default.** A fresh connection can explore and query but not change anything. Write access requires a credential with write scopes. See [Permissions and Safety](/api-and-integrations/api-access/mcp-integration/mcp-permissions-and-safety.md).
* **Confirmation before consequential actions.** Deleting a twin, editing the ontology, or starting a training run does nothing until the assistant asks you and you agree.
* **Long jobs run in the background.** Imports, discovery, and training take minutes; the assistant starts them and polls instead of blocking the conversation.
* **Guided skills.** Packaged workflows for the involved procedures, loaded by the assistant on its own or invoked by you. See [Guided Skills](/api-and-integrations/api-access/mcp-integration/mcp-skills.md).
* **Interactive results.** Tables, charts, and causal graphs render as live views in clients that support MCP Apps; elsewhere, images and text with deep links back to the platform. See [Interactive Results](/api-and-integrations/api-access/mcp-integration/mcp-interactive-results.md).

## Connection methods

Both are configured from the **API** page of your Organization home (see [API Access](/api-and-integrations/api-access.md)):

* **Remote OAuth (recommended).** The config is just a URL. Sign in through the browser and approve what the assistant may reach. No secrets stored.
* **API key (legacy).** A locally run bridge authenticated by a static `pk_` key.

Setup steps for each client: [Connecting AI Clients](/api-and-integrations/api-access/mcp-integration/mcp-connecting-clients.md).

## In this section

| Page                                                                                                     | Covers                                                                                                              |
| -------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| [Connecting AI Clients](/api-and-integrations/api-access/mcp-integration/mcp-connecting-clients.md)      | Setup for Claude Code, Claude Desktop, Claude on the web, and Cursor; the consent screen; revoking; troubleshooting |
| [What You Can Do](/api-and-integrations/api-access/mcp-integration/mcp-capabilities.md)                  | Capabilities by goal, plus the server's terminology                                                                 |
| [Guided Skills](/api-and-integrations/api-access/mcp-integration/mcp-skills.md)                          | The packaged workflows and how to invoke them as slash commands                                                     |
| [Permissions and Safety](/api-and-integrations/api-access/mcp-integration/mcp-permissions-and-safety.md) | Scopes, the confirmation gate, background jobs, rate limits                                                         |
| [Interactive Results](/api-and-integrations/api-access/mcp-integration/mcp-interactive-results.md)       | Live tables, charts, and causal graph views inside your AI client                                                   |