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

# Connect with MCP

> Connect other agent frameworks using their remote MCP support.

If your framework already supports remote MCP, you can connect directly without installing `@agentmuxer/sdk`.

| Setting        | Value                                                              |
| -------------- | ------------------------------------------------------------------ |
| Endpoint       | `https://mcp.agentmuxer.com/mcp`                                   |
| Transport      | Streamable HTTP                                                    |
| Authentication | `Authorization: Bearer <application-key>`                          |
| Credentials    | An [AgentMuxer application key](/sdk/applications); no OAuth login |

<Note>
  The examples below show the connection settings to add to an existing agent. For complete runnable examples, use the [SDK guides](/quickstart). Check [compatibility](/reference/compatibility) for supported integrations and runtime requirements.
</Note>

## Configure your framework

Read `AGENTMUXER_API_KEY` from your server's environment and reject startup if it is missing:

```typescript theme={null}
const apiKey = process.env.AGENTMUXER_API_KEY;
if (!apiKey) throw new Error("Set AGENTMUXER_API_KEY");
```

<Tabs>
  <Tab title="OpenAI Responses">
    Add this item to the Responses API's `tools` array:

    ```typescript theme={null}
    {
      type: "mcp",
      server_label: "agentmuxer",
      server_url: "https://mcp.agentmuxer.com/mcp",
      authorization: apiKey,
      require_approval: "never",
    }
    ```

    OpenAI receives the application key and connects to AgentMuxer. `require_approval` controls OpenAI's approval step; AgentMuxer's payment policy still applies. See [OpenAI remote MCP](https://developers.openai.com/api/docs/guides/tools-remote-mcp).
  </Tab>

  <Tab title="Anthropic Messages">
    Add these fields to a beta Messages request:

    ```typescript theme={null}
    {
      betas: ["mcp-client-2025-11-20"],
      mcp_servers: [{
        type: "url",
        name: "agentmuxer",
        url: "https://mcp.agentmuxer.com/mcp",
        authorization_token: apiKey,
      }],
      tools: [{
        type: "mcp_toolset",
        mcp_server_name: "agentmuxer",
      }],
    }
    ```

    Anthropic receives the application key and connects to AgentMuxer. The connector is beta and supports tool calls; check its [current requirements](https://platform.claude.com/docs/en/agents-and-tools/mcp-connector) before deploying.
  </Tab>

  <Tab title="Mastra">
    Install `@mastra/mcp`, then configure its client:

    ```typescript theme={null}
    import { MCPClient } from "@mastra/mcp";

    const client = new MCPClient({
      servers: {
        agentmuxer: {
          url: new URL("https://mcp.agentmuxer.com/mcp"),
          requestInit: {
            headers: { Authorization: `Bearer ${apiKey}` },
          },
        },
      },
    });
    ```

    Pass `await client.listTools()` to your agent's tools configuration. Call `await client.disconnect()` when the run finishes, including on errors. See [Mastra MCP](https://mastra.ai/docs/connections/mcp).
  </Tab>

  <Tab title="LangChain / LangGraph">
    Install `@langchain/mcp-adapters`, then configure its client:

    ```typescript theme={null}
    import { MultiServerMCPClient } from "@langchain/mcp-adapters";

    const client = new MultiServerMCPClient({
      agentmuxer: {
        transport: "http",
        url: "https://mcp.agentmuxer.com/mcp",
        headers: { Authorization: `Bearer ${apiKey}` },
      },
    });
    ```

    Pass `await client.getTools()` to your agent's tools configuration. Release the client when the run ends using the lifecycle API for your installed version. See [LangChain MCP](https://docs.langchain.com/oss/javascript/langchain/mcp).
  </Tab>
</Tabs>

## Permissions and spending

These connections expose AgentMuxer's tools, including purchases. Your framework owns model configuration, tool selection, permissions, and connection lifecycle. AgentMuxer enforces the account balance, application spending limit, and payment-confirmation policy.

Before using a headless workflow, confirm that it can handle your account's [payment approval requirements](/guides/billing). Model-provider charges are separate. Keep application keys out of browser bundles, prompts, tool arguments, and logs.

<CardGroup cols={2}>
  <Card title="Use an SDK helper" icon="code" href="/reference/sdk">
    See the OpenAI, Claude, and Vercel integration options.
  </Card>

  <Card title="Troubleshooting" icon="wrench" href="/reference/troubleshooting">
    Resolve authentication, tool, and purchase errors.
  </Card>
</CardGroup>