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

# Other MCP clients

> Other remote MCP clients, local stdio, API Keys, and tool verification

## Choose an MCP connection

| Connection | Install or connect | Authentication | For |
| - | - | - | - |
| Remote MCP | [https://mcp.evolink.ai/mcp](https://mcp.evolink.ai/mcp) | Browser OAuth | Web chat and native MCP clients |
| Local MCP | npx starts @evolinkai/mcp | Platform API key | stdio clients and allowed local uploads |

Remote MCP does not require the local MCP package. Installing CLI does not register a native MCP server. Installing a skill does not sign in or connect tools.

## Choose a dedicated guide

Use the separate pages for [Claude](/docs/en/mcp/claude), [Codex](/docs/en/mcp/codex), [Claude Code](/docs/en/mcp/claude-code), [ChatGPT](/docs/en/mcp/chatgpt), [Cursor](/docs/en/mcp/cursor), [OpenClaw](/docs/en/mcp/openclaw), and [Hermes](/docs/en/mcp/hermes-agent). This page covers generic remote MCP and local stdio connections for additional clients.

<span id="cli-install" />

## Separate CLI guides for terminal agents

For terminals, scripts, Gemini CLI, OpenCode, and GitHub Copilot CLI, follow [EvoLink CLI setup](/docs/en/cli/setup#cli-install) for installation, sign-in, skill targets, and free verification. Agents that execute commands can use CLI; configure native MCP with the remote or local instructions below.

<span id="ssh-login" />

### Linux and SSH sign-in

When CLI runs remotely but your browser is local, follow [SSH callback forwarding](/docs/en/cli/setup#ssh-login) after meeting Linux secure credential storage requirements. Remote MCP callbacks are managed by the host and do not use the CLI forwarding steps.

<span id="local-mcp" />

## Local stdio MCP with an API Key

A stdio-only client can launch a local MCP process. It needs **Node.js 18+** and a personal API Key, rather than remote MCP browser OAuth.

<Steps>
  <Step title="Prepare Node and a dedicated Key">
    ```bash theme={null}
    node --version
    npx --version
    ```

    Both commands must work with Node 18 or newer. Create a dedicated personal Key with limits in [API Keys](https://evolink.ai/dashboard/keys). Supply `EVOLINK_API_KEY` through the client’s safe environment or secret settings, not agent chat, shared screenshots, or a repository.
  </Step>

  <Step title="Configure the local launcher">
    Choose local/stdio in the client’s MCP settings:

    ```text theme={null}
    command: npx
    args: -y @evolinkai/mcp@latest
    ```

    On first launch, npx retrieves the package. The host communicates over stdio and the local process calls the platform API. Ensure the host can find npx; a GUI application may have a different environment from your terminal. `YOUR_API_KEY` below is a placeholder for private local configuration. Prefer environment inheritance when supported.

    ```json theme={null}
    {
      "mcpServers": {
        "evolink": {
          "command": "npx",
          "args": ["-y", "@evolinkai/mcp@latest"],
          "env": {"EVOLINK_API_KEY": "YOUR_API_KEY"}
        }
      }
    }
    ```
  </Step>

  <Step title="Allow directories when local files are needed">
    For local-path uploads, set `EVOLINK_UPLOAD_ALLOWED_DIRS` to allowed absolute directories in the process environment. Separate multiple directories with `:` on macOS/Linux or `;` on Windows. Local-path uploads are disabled without this setting. Remote MCP never accepts `file_path`.

    This controls local file access, not model format, duration, or size limits. URL/base64 usage does not need local directory access.
  </Step>

  <Step title="Reload and verify free tools">
    Save configuration, enable EvoLink, and reload or open a new session as your host requires. Ask it to call `check_balance` and `search_models` and confirm real results. Starting a process or listing tools alone is not successful account verification.

    Local stdio does not provide remote single-use upload tools `prepare_upload` / `get_upload`. Inspect actual `tools/list`. The personal Key has its own limits, not OAuth Key limits.
  </Step>
</Steps>

### Claude Code and Codex configuration

With the Key already provided through the safe environment, Claude Code can register using this command instead of the JSON example:

```bash theme={null}
claude mcp add --scope user evolink -- npx -y @evolinkai/mcp@latest
```

Codex can merge this section to inherit variables from its startup environment:

```toml theme={null}
[mcp_servers.evolink]
command = "npx"
args = ["-y", "@evolinkai/mcp@latest"]
env_vars = ["EVOLINK_API_KEY", "EVOLINK_UPLOAD_ALLOWED_DIRS"]
```

Reload after updating and inspect the actual version. `@latest` does not replace an already running process. See [central troubleshooting](/docs/en/mcp/billing#mcp-connection) for startup, Key, and local-path failures.

<span id="remote-mcp" />

## Other remote MCP clients

The client needs Streamable HTTP, OAuth, and browser callbacks. A client limited to static Keys or stdio should use local MCP above.

<Steps>
  <Step title="Add the server">
    Create an MCP connection named EvoLink with Streamable HTTP and this URL:

    ```text theme={null}
    https://mcp.evolink.ai/mcp
    ```

    Do not enter a documentation URL, platform `/v1` URL, or npx command.
  </Step>

  <Step title="Complete OAuth and enable tools">
    Choose OAuth, use the client’s sign-in action, approve in the browser, and return. Confirm the connection and tools are enabled. A static API Key is not this connection’s OAuth token.
  </Step>

  <Step title="Verify for free">
    ```text theme={null}
    Use EvoLink to check my balance and search available image models. Do not generate anything yet.
    ```

    After actual balance and model results return, follow the [first media workflow](/docs/en/mcp/overview#workflows). A tool list without successful calls still needs authorization troubleshooting.
  </Step>
</Steps>

<span id="skills" />

## Skills and plugin resources

The bundled evolink-cli skill guides command execution; repository evolink-mcp guides native tools. See [agent resources](/docs/en/cli-mcp/agent-resources#skills) for sources, directories, and updates, and the [plugin guide](/docs/en/mcp/plugins) for packaged remote configuration.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.