MorphLLM OpenCode Plugin

An OpenCode plugin that integrates MorphLLM's intelligent model routing and powerful MCP tools into your AI coding workflow.

• GitHub: https://github.com/VitoLin/opencode-morphllm
• NPM: https://www.npmjs.com/package/opencode-morphllm

Automatically classifies your prompts by complexity (easy/medium/hard) and routes them to the most cost-effective and appropriate model. Save money on simple tasks while using powerful models only when needed.

Adds MorphLLM's specialized MCP tools to OpenCode:

• edit_file - Fast, precise file editing tool optimized for code modifications
• warpgrep_codebase_search - Advanced codebase search with contextual understanding

Optional mode that maintains model consistency within a session, reducing costs through prompt caching.

Add the plugin to your OpenCode configuration at ~/.config/opencode/opencode.json:

{
  "plugins": ["opencode-morphllm"]
}

Create a configuration file at ~/.config/opencode/morph.json or for repo level configs, you can create the config at your repo root at .opencode/morph.json.

Example:

{
  "MORPH_API_KEY": "your_morph_api_key_here",
  "MORPH_ROUTER_CONFIGS": {
    "MORPH_MODEL_EASY": "github-copilot/gpt-5-mini",
    "MORPH_MODEL_MEDIUM": "opencode/minimax-m2.1-free",
    "MORPH_MODEL_HARD": "github-copilot/gemini-2.5-pro",
    "MORPH_ROUTER_ENABLED": true,
    "MORPH_ROUTER_PROMPT_CACHING_AWARE": false
  }
}

Find available models at models.dev.

Note about router enablement

• The router will be automatically disabled if none of the model slots (MORPH_MODEL_EASY, MORPH_MODEL_MEDIUM, MORPH_MODEL_HARD) are configured (i.e., all are empty strings). This prevents the router from being active when there are no target models configured. An explicit MORPH_ROUTER_ENABLED: false in your config will also disable the router even if models are present.

The MORPH_SYSTEM_MESSAGE setting allows you to customize or override the default system message that MorphLLM appends to OpenCode's system prompt. This message guides the AI on using Morph's MCP tools effectively.

Default message:

You **MUST** consider using morph_mcp. For editing files, consider using morph_mcp_edit_file. For searching the code base, consider using warpgrep_codebase_search

When to customize:

• You want to emphasize different tool preferences for your workflow
• You need to add project-specific conventions or guidelines
• You want to disable the default guidance entirely (set to empty string "")

Example:

{
  "MORPH_API_KEY": "your_morph_api_key_here",
  "MORPH_SYSTEM_MESSAGE": "Your custom system message here"
}

Models are specified as provider/model-id. Examples:

• github-copilot/gpt-5-mini
• github-copilot/gemini-2.5-pro
• opencode/minimax-m2.1-free

For backward compatibility, you can also use flat configuration:

{
  "MORPH_API_KEY": "your_key",
  "MORPH_MODEL_EASY": "github-copilot/gpt-5-mini",
  "MORPH_MODEL_MEDIUM": "opencode/minimax-m2.1-free",
  "MORPH_MODEL_HARD": "github-copilot/gemini-2.5-pro",
  "MORPH_ROUTER_ENABLED": true
}

You can also configure per-project settings by creating .opencode/morph.json in your project root. Project settings override user settings.

my-project/
├── .opencode/
│   └── morph.json      # Project-specific Morph config
└── src/

Both .json and .jsonc (with comments) formats are supported.

When you send a prompt, the plugin:

1. Classifies the prompt complexity using MorphLLM's classification engine
2. Selects the appropriate model based on your configuration
3. Routes the request to the chosen model

Example routing:

• "Format this JSON" → Easy → MORPH_MODEL_EASY (cheaper, faster)
• "Add a React component" → Medium → MORPH_MODEL_MEDIUM
• "Debug this race condition" → Hard → MORPH_MODEL_HARD (powerful)

When MORPH_ROUTER_PROMPT_CACHING_AWARE is enabled:

• The first prompt in a session determines the model for all subsequent prompts
• Reduces costs through prompt caching with compatible providers
• Best for long conversations on similar topics

See DEVELOPMENT.md for setup instructions and development notes.

Ensure the plugin is properly listed in your OpenCode config:

{
  "plugins": ["opencode-morphllm"]
}

Check that:

1. MORPH_API_KEY is set correctly
2. MORPH_ROUTER_ENABLED is not set to false
3. At least one model is configured

Visit models.dev to browse available models and their IDs.

MIT