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

# OpenClaw

> Use the latest MiniMax M-series models for anything you can think of in OpenClaw.

<div style={{background:"#fffbeb",borderLeft:"4px solid #d97706",padding:"12px 16px",borderRadius:"6px",margin:"16px 0"}}>[**OpenClaw**](https://github.com/openclaw/openclaw) is a personal AI assistant running locally, integrating with messaging platforms for control.</div>

## Prerequisites

* A MiniMax [Subscription Key](https://platform.minimax.io/user-center/payment/token-plan) with Token Plan or Credits access
* A computer with terminal access (macOS, Linux, or Windows with WSL)

***

## Install and Configure OpenClaw

<Steps>
  <Step title="Run Installation Command">
    Run the installation command in your terminal:

    ```bash theme={null}
    curl -fsSL https://openclaw.ai/install.sh | bash
    ```

    You'll see this entry page:

    <img src="https://file.cdn.minimax.io/public/0fc6ba49-4324-4d69-ae58-56e4f5eac647.png" alt="Entry page" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} />
  </Step>

  <Step title="Start Setup">
    Select **"Yes"** .
  </Step>

  <Step title="Select QuickStart Mode">
    Select **"QuickStart"** to use the guided setup.
  </Step>

  <Step title="Select Model Provider">
    Select **"MiniMax"** as the model provider.
  </Step>

  <Step title="Select Auth Method">
    Select **"MiniMax Global — OAuth (minimax.io)"** as the authentication method.
  </Step>

  <Step title="Sign In to MiniMax">
    You'll be sent to a browser window to sign in to or sign up for your MiniMax API Platform account.
  </Step>

  <Step title="Authorize OpenClaw">
    Select **"Authorize"**, after which you can return to the terminal.

    <img src="https://mintcdn.com/minimax-cac98058/Z6t8NKWdGWkRU7_m/images/openclaw-authorize.png?fit=max&auto=format&n=Z6t8NKWdGWkRU7_m&q=85&s=8f6f82d5c7e0ed9288ff7c055542d43f" alt="Authorize" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} width="1280" height="800" data-path="images/openclaw-authorize.png" />
  </Step>

  <Step title="Confirm Model Selection">
    A model picker will appear with the available models already selected. Press **Enter** to continue.
  </Step>

  <Step title="Select a Messaging Channel">
    Back in the terminal, select a messaging channel to connect. OpenClaw supports a range of platforms including Telegram, WhatsApp, Discord, iMessage, and more.

    <img src="https://file.cdn.minimax.io/public/6307996b-250d-45be-b004-68b1d3ddd396.png" alt="Select channel" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} />
  </Step>

  <Step title="Confirm Installation">
    Follow the prompts for your chosen channel, then select **"Yes"** when asked to confirm.

    <img src="https://file.cdn.minimax.io/public/b1d04c91-ab5f-494a-a798-ce9484eb71f2.png" alt="Confirm" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} />
  </Step>

  <Step title="Select Skills">
    Select any additional skills you wish to install (press space to select, up and down arrows to navigate).

    <img src="https://file.cdn.minimax.io/public/560c2837-c62c-4685-acd9-26fc0e0648db.png" alt="Select skills" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} />
  </Step>

  <Step title="Configure API Keys">
    Configure additional API Keys for tools that OpenClaw can use (optional). Select **"Skip for now"** if you don't need additional tools.

    <img src="https://file.cdn.minimax.io/public/668de94c-6ab0-4e61-8a1b-7312a0c2357d.png" alt="Configure API Keys" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} />
  </Step>

  <Step title="Open Web UI">
    Select **"Open the Web UI"** to access the OpenClaw dashboard.

    <img src="https://file.cdn.minimax.io/public/2ae52543-decb-4269-8d4d-475577e3edfe.png" alt="Open Web UI" style={{borderRadius: '8px', marginTop: '12px', maxWidth: '100%'}} />
  </Step>
</Steps>

***

## Toggle thinking

Toggle thinking at runtime with `/think`: `off` turns it off, `adaptive` (the default) lets M3 decide when to think.

***

## Advanced Capabilities

### Image Understanding

<Warning>
  When logged in via OAuth, OpenClaw's `image` tool is **automatically wired** to MiniMax's [Image Understanding MCP server](/token-plan/mcp-guide), giving your agent image understanding capabilities out of the box — no extra setup needed.

  If you authenticate with an **API Key manually**, or your OpenClaw build doesn't auto-include the Image Understanding MCP, pick one of the methods below to add image understanding to your agent.
</Warning>

<Tabs>
  <Tab title="Method 1 (Recommended): MiniMax CLI">
    [MiniMax CLI](https://github.com/MiniMax-AI/cli) (the `mmx` command) is the official multi-modal command-line tool. Once installed, OpenClaw can invoke `mmx vision` for image understanding — and you get **the same CLI for language, image, video, speech, music, and search** (7 multi-modal capabilities).

    <Steps>
      <Step title="Prepare Environment">
        Make sure Node.js 18+ is available:

        ```bash theme={null}
        # Check Node.js version (should be >= 18)
        node -v

        # If not installed, install LTS via nvm
        curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash
        nvm install --lts
        ```
      </Step>

      <Step title="Install MiniMax CLI">
        Run the following command in your terminal for a global install:

        ```bash theme={null}
        npm install -g mmx-cli
        ```

        ⭐️ Or hand install, login, and SKILL setup to OpenClaw in one go (replace `sk-xxxxx` with your actual key):

        ```plaintext theme={null}
        Please set up MiniMax CLI (https://github.com/MiniMax-AI/cli) for me with these three steps:

        1. Install CLI globally: run `npm install -g mmx-cli`, then verify with `mmx --version`
        2. Sign in with API Key: run `mmx auth login --api-key sk-xxxxx`
        3. Install the official SKILL: run `npx skills add MiniMax-AI/cli -y -g`

        After that, please run `mmx quota` to check my Token Plan balance and confirm everything is wired up.
        ```
      </Step>

      <Step title="Login with API Key">
        Authenticate with your API Key (replace `sk-xxxxx` with your own Key):

        ```bash theme={null}
        mmx auth login --api-key sk-xxxxx
        ```

        The latest mmx-cli auto-detects your service region from the Key — no manual region setup is usually needed.

        <Note>
          The region follows the platform where your API service was purchased: `cn` for the [MiniMax China Token Plan](https://platform.minimaxi.com/subscribe/token-plan), or `global` for the [MiniMax International Token Plan](https://platform.minimax.io/subscribe/token-plan).
        </Note>

        <Warning>
          If you hit a 401 after logging in, the region likely didn't auto-match. Set it manually:

          ```bash theme={null}
          mmx config set --key region --value cn       # China API service
          mmx config set --key region --value global   # International API service
          ```

          Run `mmx auth status` to confirm the region matches the platform where your subscription was purchased.
        </Warning>

        Once logged in, run `mmx quota` to view your Token Plan balance and confirm the setup is working.
      </Step>

      <Step title="Install SKILL (Optional, Recommended for Agent Users)">
        If you'll be invoking `mmx` from inside OpenClaw, we recommend adding the official SKILL.md so the agent makes better tool calls without having to read `--help` on the fly:

        ```bash theme={null}
        npx skills add MiniMax-AI/cli -y -g
        ```

        <Tip>
          The SKILL is auto-symlinked into `~/.openclaw/skills/`, and OpenClaw will pick it up on the next launch.
        </Tip>
      </Step>

      <Step title="Tell OpenClaw to Use mmx vision">
        Tell OpenClaw to prefer `mmx vision` for any future image understanding:

        ```plaintext theme={null}
        Going forward, prefer the mmx vision tool for image understanding,
        invoke it via `mmx vision describe --image <image path or URL>` and parse the result.
        ```
      </Step>

      <Step title="Verify">
        Send an image in OpenClaw and check that it auto-invokes `mmx vision describe --image <path>` and returns a correct description.
      </Step>
    </Steps>
  </Tab>

  <Tab title="Method 2: Token Plan MCP">
    Wire up the [Token Plan MCP](/token-plan/mcp-guide) to use the `understand_image` tool.

    <Steps>
      <Step title="Prepare Environment">
        Make sure uv (the Python package manager) is available:

        ```bash theme={null}
        # Check
        which uv

        # If not installed, via pip
        pip install uv

        # Or via curl
        curl -LsSf https://astral.sh/uv/install.sh | sh
        ```
      </Step>

      <Step title="Install and Configure MCP">
        For manual setup, follow the [Token Plan MCP guide](/token-plan/mcp-guide). Core steps:

        1. Install the `mcporter` skill (an MCP server manager)
        2. Configure `minimax-coding-plan-mcp` per the official guide
        3. Set the `MINIMAX_API_KEY` environment variable (Token Plan API Key starts with `sk-cp`):

        ```bash theme={null}
        export MINIMAX_API_KEY="sk-..."
        ```

        ⭐️ Or hand the entire flow to OpenClaw with this prompt (**replace `sk-xxxxxxx` with your actual API Key**):

        ```plaintext theme={null}
        First install the mcporter skill, then follow https://platform.minimax.io/docs/token-plan/mcp-guide to configure the MCP. My MiniMax API Key is sk-xxxxxxx. Finally use the mcporter skill to wire it up.
        ```
      </Step>

      <Step title="Tell OpenClaw to Use understand_image">
        Tell OpenClaw to prefer `understand_image` for image understanding:

        ```plaintext theme={null}
        Going forward, use the understand_image tool from minimax-coding-plan-mcp for any image understanding.
        ```
      </Step>

      <Step title="Verify">
        Send an image in OpenClaw and check it auto-invokes the `understand_image` tool and returns a correct description.
      </Step>
    </Steps>
  </Tab>
</Tabs>

***

### Web Search

OpenClaw doesn't ship with web search out of the box. Add it via one of the methods below (Token Plan subscribers, API Key starts with `sk-cp`).

<Tabs>
  <Tab title="Method 1 (Recommended): MiniMax CLI">
    `mmx search` is the web-search command from MiniMax CLI. No separate MCP server process — OpenClaw invokes it directly via Bash.

    <Steps>
      <Step title="Prepare Environment">
        Make sure Node.js 18+ is available:

        ```bash theme={null}
        # Check Node.js version (should be >= 18)
        node -v

        # If not installed, install LTS via nvm
        curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.7/install.sh | bash
        nvm install --lts
        ```
      </Step>

      <Step title="Install MiniMax CLI">
        Run the following command in your terminal for a global install:

        ```bash theme={null}
        npm install -g mmx-cli
        ```

        ⭐️ Or hand install, login, and SKILL setup to OpenClaw in one go (replace `sk-xxxxx` with your actual key):

        ```plaintext theme={null}
        Please set up MiniMax CLI (https://github.com/MiniMax-AI/cli) for me with these three steps:

        1. Install CLI globally: run `npm install -g mmx-cli`, then verify with `mmx --version`
        2. Sign in with API Key: run `mmx auth login --api-key sk-xxxxx`
        3. Install the official SKILL: run `npx skills add MiniMax-AI/cli -y -g`

        After that, please run `mmx quota` to check my Token Plan balance and confirm everything is wired up.
        ```
      </Step>

      <Step title="Login with API Key">
        Authenticate with your API Key (replace `sk-xxxxx` with your own Key):

        ```bash theme={null}
        mmx auth login --api-key sk-xxxxx
        ```

        The latest mmx-cli auto-detects your service region from the Key — no manual region setup is usually needed.

        <Note>
          The region follows the platform where your API service was purchased: `cn` for the [MiniMax China Token Plan](https://platform.minimaxi.com/subscribe/token-plan), or `global` for the [MiniMax International Token Plan](https://platform.minimax.io/subscribe/token-plan).
        </Note>

        <Warning>
          If you hit a 401 after logging in, the region likely didn't auto-match. Set it manually:

          ```bash theme={null}
          mmx config set --key region --value cn       # China API service
          mmx config set --key region --value global   # International API service
          ```

          Run `mmx auth status` to confirm the region matches the platform where your subscription was purchased.
        </Warning>

        Once logged in, run `mmx quota` to view your Token Plan balance and confirm the setup is working.
      </Step>

      <Step title="Install SKILL (Optional, Recommended for Agent Users)">
        If you'll be invoking `mmx` from inside OpenClaw, we recommend adding the official SKILL.md so the agent makes better tool calls without having to read `--help` on the fly:

        ```bash theme={null}
        npx skills add MiniMax-AI/cli -y -g
        ```

        <Tip>
          The SKILL is auto-symlinked into `~/.openclaw/skills/`, and OpenClaw will pick it up on the next launch.
        </Tip>
      </Step>

      <Step title="Tell OpenClaw to Use mmx search">
        `mmx search` offers two invocation forms:

        ```bash theme={null}
        # Simple call (human-friendly, readable text output)
        mmx search "latest MiniMax AI updates"

        # Structured call (Agent-friendly)
        mmx search query --q "MiniMax M2.7 release" --output json
        ```

        Example prompt for OpenClaw:

        ```plaintext theme={null}
        Use the mmx search tool to search "OpenClaw multi-Agent orchestration",
        invoke it via `mmx search query --q "..." --output json` to get structured results,
        then summarize the top 3 hits' titles, links, and key snippets.
        ```

        OpenClaw will run `mmx search query --q "..." --output json` via the Bash tool and hand the stdout JSON to the model directly — no MCP server process needed.
      </Step>

      <Step title="Memorize">
        ```plaintext theme={null}
        Going forward, prefer the mmx search tool for web search,
        invoke it via `mmx search query --q "..." --output json` and parse the JSON.
        ```
      </Step>
    </Steps>
  </Tab>

  <Tab title="Method 2: Token Plan MCP">
    Wire up the [Token Plan MCP](/token-plan/mcp-guide) to use the `web_search` tool.

    <Steps>
      <Step title="Prepare Environment">
        Make sure uv (the Python package manager) is available:

        ```bash theme={null}
        # Check
        which uv

        # If not installed, via pip
        pip install uv

        # Or via curl
        curl -LsSf https://astral.sh/uv/install.sh | sh
        ```
      </Step>

      <Step title="Install and Configure MCP">
        For manual setup, follow the [Token Plan MCP guide](/token-plan/mcp-guide). Core steps:

        1. Install the `mcporter` skill (an MCP server manager)
        2. Configure `minimax-coding-plan-mcp` per the official guide
        3. Set the `MINIMAX_API_KEY` environment variable (Token Plan API Key starts with `sk-cp`):

        ```bash theme={null}
        export MINIMAX_API_KEY="sk-..."
        ```

        ⭐️ Or hand the entire flow to OpenClaw with this prompt (**replace `sk-xxxxxxx` with your actual API Key**):

        ```plaintext theme={null}
        First install the mcporter skill, then follow https://platform.minimax.io/docs/token-plan/mcp-guide to configure the MCP. My MiniMax API Key is sk-xxxxxxx. Finally use the mcporter skill to wire it up.
        ```
      </Step>

      <Step title="Tell OpenClaw to Use web_search">
        Tell OpenClaw to prefer `web_search`:

        ```plaintext theme={null}
        Going forward, use the web_search tool from minimax-coding-plan-mcp for any web search.
        ```
      </Step>

      <Step title="Verify">
        Ask OpenClaw something requiring up-to-date info, and check it auto-invokes the `web_search` tool and returns search results.
      </Step>
    </Steps>
  </Tab>
</Tabs>

<Tip>
  For a full step-by-step walkthrough of connecting OpenClaw to Telegram (including bot creation and pairing), see the [OpenClaw Cookbook guide](/solutions/openclaw).
</Tip>