---
updatedAt: 2026-10-08T21:36:10.000Z
agentTools:
  projectIndex: https://docs.superwise.ai/llms.txt
---

# MCP

How to connect AI editors to the SUPERWISE docs and Python SDK MCP servers

SUPERWISE® publishes two Model Context Protocol (MCP) servers for reading the SUPERWISE documentation inside an AI assistant or code editor, such as Cursor, Windsurf, Claude Desktop, Claude Code, or Codex. Add them, and the assistant answers questions about SUPERWISE from these docs and the Python SDK reference instead of from memory.

## What is MCP?

Model Context Protocol (MCP) is an open standard that lets an AI assistant read from outside sources and call outside tools. The SUPERWISE MCP servers give the assistant:

* **Documentation search** across the user guides, the REST API reference, and the Python SDK reference
* **Code examples** drawn from those pages, for HTTP calls and for the `superwise_api` package

## Available servers

| Server                                | URL                                     | Best for                                                                                              |
| ------------------------------------- | --------------------------------------- | ----------------------------------------------------------------------------------------------------- |
| **SUPERWISE Docs: API & User Guides** | `https://docs.superwise.ai/mcp`         | Searching the user guides and the REST API reference, and writing HTTP-based integration code.        |
| **SUPERWISE Python SDK Reference**    | `https://mcp.sdk.docs.superwise.ai/mcp` | Writing Python code with the `superwise_api` package: classes, methods, schemas, Quickstart examples. |

Both servers speak streamable HTTP MCP and need no sign-in to read the public docs.
The docs server can also send a REST API request if you add your own SUPERWISE
credentials in your MCP client. Without them, both servers read documentation only.

### Which one should I use?

Most editors can register multiple MCP servers at once — keep both enabled and
your editor will pick the right one per question. A rough guide:

* Writing Python (`import superwise_api`, `pip install superwise-api`, class
  and schema reference) → **Python SDK Reference** server.
* Calling REST endpoints, generating curl examples, or asking about product
  concepts (policies, dashboards, agents, guardrails) → **API & User Guides** server.

## Setup

### SUPERWISE Docs: REST API & User Guides

SUPERWISE hosts a remote MCP server at `https://docs.superwise.ai/mcp`. Add it to your AI tool with one of the configurations below. Each one names the server `superwise-docs`; any name works.

<Tabs>
  <Tab title="Cursor">
    **Add to`~/.cursor/mcp.json`:**

    ```json
    {
      "mcpServers": {
        "superwise-docs": {
          "url": "https://docs.superwise.ai/mcp"
        }
      }
    }
    ```
  </Tab>

  <Tab title="Windsurf">
    **Add to`~/.codeium/windsurf/mcp_config.json`:**

    ```json
    {
      "mcpServers": {
        "superwise-docs": {
          "url": "https://docs.superwise.ai/mcp"
        }
      }
    }
    ```
  </Tab>

  <Tab title="Claude Desktop">
    Claude Desktop only launches stdio MCP subprocesses, so remote HTTP servers
    are bridged through [`mcp-remote`](https://www.npmjs.com/package/mcp-remote).
    Requires Node.js (`npx`) on your PATH.

    **Add to`claude_desktop_config.json`:**

    ```json
    {
      "mcpServers": {
        "superwise-docs": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://docs.superwise.ai/mcp"]
        }
      }
    }
    ```

    Fully quit (Cmd+Q / Alt+F4) and relaunch Claude Desktop after editing.
  </Tab>

  <Tab title="Codex">
    Codex CLI configures MCP servers in TOML. Remote HTTP servers are bridged
    through [`mcp-remote`](https://www.npmjs.com/package/mcp-remote). Requires
    Node.js (`npx`) on your PATH.

    **Add to`~/.codex/config.toml`:**

    ```toml
    [mcp_servers.superwise-docs]
    command = "npx"
    args = ["-y", "mcp-remote", "https://docs.superwise.ai/mcp"]
    ```
  </Tab>
</Tabs>

### SUPERWISE Python SDK Reference

<Tabs>
  <Tab title="Cursor">
    **Add to`~/.cursor/mcp.json`:**

    ```json
    {
      "mcpServers": {
        "superwise-sdk-docs": {
          "url": "https://mcp.sdk.docs.superwise.ai/mcp"
        }
      }
    }
    ```
  </Tab>

  <Tab title="Windsurf">
    **Add to`~/.codeium/windsurf/mcp_config.json`:**

    ```json
    {
      "mcpServers": {
        "superwise-sdk-docs": {
          "url": "https://mcp.sdk.docs.superwise.ai/mcp"
        }
      }
    }
    ```
  </Tab>

  <Tab title="Claude Desktop">
    Claude Desktop only launches stdio MCP subprocesses, so remote HTTP servers
    are bridged through [`mcp-remote`](https://www.npmjs.com/package/mcp-remote).
    Requires Node.js (`npx`) on your PATH.

    **Add to`claude_desktop_config.json`:**

    ```json
    {
      "mcpServers": {
        "superwise-sdk-docs": {
          "command": "npx",
          "args": ["-y", "mcp-remote", "https://mcp.sdk.docs.superwise.ai/mcp"]
        }
      }
    }
    ```

    Fully quit (Cmd+Q / Alt+F4) and relaunch Claude Desktop after editing.
  </Tab>

  <Tab title="Codex">
    Codex CLI configures MCP servers in TOML. Remote HTTP servers are bridged
    through [`mcp-remote`](https://www.npmjs.com/package/mcp-remote). Requires
    Node.js (`npx`) on your PATH.

    **Add to`~/.codex/config.toml`:**

    ```toml
    [mcp_servers.superwise-sdk-docs]
    command = "npx"
    args = ["-y", "mcp-remote", "https://mcp.sdk.docs.superwise.ai/mcp"]
    ```
  </Tab>
</Tabs>

## Testing Your MCP Setup

Once configured, you can test your MCP server connection:

1. **Open your AI editor** (Cursor, Windsurf, etc.)
2. **Start a new chat** with the AI assistant
3. **Ask about SUPERWISE Docs or SDK** - try questions like:
   * *"Using the SUPERWISE Python SDK, show me how to create a Policy."*
   * *"List the available Dashboard methods in `superwise_api`."*
   * *"Using the SUPERWISE REST API, how do I POST a new agent?"*

The assistant should now answer from the SUPERWISE documentation through the MCP servers.