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

# MCP installation by client

> Configure DreamLayer for Claude Code, Codex, Cursor, Hermes, OpenClaw, or another stdio MCP client.

# MCP installation by client

## What this lets you do

Connect Claude Code, Codex, Cursor, Hermes, OpenClaw, or another stdio MCP client to DreamLayer image tools.

## What you need

* A DreamLayer clone
* Python and `uv`
* One supported MCP client

## Steps

### 1. Install the command

```bash theme={null}
uv tool install ./packages/mcp-server
```

### 2. Configure Claude Code

```bash theme={null}
claude mcp add dreamlayer -- dreamlayer-mcp
```

### 3. Configure Codex

```bash theme={null}
codex mcp add dreamlayer -- dreamlayer-mcp
```

### 4. Configure Cursor, Hermes, OpenClaw, or a generic client

Add a local stdio server to the client's MCP configuration:

```json theme={null}
{
  "mcpServers": {
    "dreamlayer": { "command": "dreamlayer-mcp" }
  }
}
```

Client configuration keys change over time. Use the client's current MCP documentation for the location of this object. DreamLayer requires only a stdio command and optional environment variables.

## Confirm it worked

Restart the client, list its MCP tools, and call the DreamLayer capabilities tool. The result must report protocol version `1`.

## Common errors

* Command not found: confirm that the `uv` tool bin directory is on the client's PATH.
* Invalid JSON: validate the client configuration and restart the client.
* JSON-RPC parse failure: keep all non-protocol output on stderr.

## Troubleshooting

Run `dreamlayer-mcp` in a terminal to confirm that the command starts, then use the current client documentation to verify its configuration file location.

## Next steps

* [Review MCP tool behavior](/mcp)
* [Use DreamLayer Agent API](/agent-api)