OpenCode Easy Vision

An OpenCode plugin that adds vision support to models that lack it. Paste images directly into the chat and ask questions, just like you would with Claude or GPT.

Important

Version 2.x requires OpenCode v2 and is published under the next tag, so install it as opencode-easy-vision@next. Plain opencode-easy-vision still installs 1.x, for OpenCode v1. To keep v1 setups working when 2.x becomes the default, pin the 1.x release:

{ "plugin": ["opencode-easy-vision@1"] }

Coming from 1.x? See Upgrading from 1.x.

Table of Contents

• The Problem
• Demo
• Setup
  • Manual Setup
  • Agent-Assisted Setup
• Usage
• Configuration
• Supported Image Formats
• Troubleshooting
• Upgrading from 1.x
• Uninstallation
• Contributing
• License
• References

The Problem

Many models, especially open-weight ones, have no vision capability. They are text-only and cannot process image input. For some models, such as MiniMax, an MCP tool can read an image through a vision service and return the analysis as text. The tool needs a local file path rather than a clipboard paste, so you would otherwise need to save each screenshot and provide its path yourself.

This plugin automates the entire workaround. It intercepts pasted images, saves them to disk, and injects the instructions the model needs to call the MCP tool with the correct path. You paste an image and ask a question. The plugin handles the rest.

Demo

Without the plugin	With the plugin
The image is silently ignored by the model.	The model analyzes the attached image correctly before answering.

Snapzy_Recording_2026-07-18_21-05-38.mov

Snapzy_Recording_2026-07-18_20-58-25.mov

Setup

Manual Setup

1. Configure an MCP image analysis tool

The plugin works with any MCP server that can analyze an image from a local path or URL and return its analysis as text. Add one to your opencode.json.

Default: MiniMax Coding Plan MCP

{
  "mcp": {
    "servers": {
      "MiniMax": {
        "type": "local",
        "command": ["uvx", "minimax-coding-plan-mcp"],
        "environment": {
          "MINIMAX_API_KEY": "{env:MINIMAX_API_KEY}",
          "MINIMAX_API_HOST": "https://api.minimax.io"
        }
      }
    }
  }
}

Alternative: openrouter-image-mcp

Routes image analysis through OpenRouter, giving you access to any vision-capable model including free ones.

{
  "mcp": {
    "servers": {
      "openrouter_image": {
        "type": "local",
        "command": ["npx", "openrouter-image-mcp"],
        "environment": {
          "OPENROUTER_API_KEY": "{env:OPENROUTER_API_KEY}",
          "OPENROUTER_MODEL": "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free"
        }
      }
    }
  }
}

Tip

nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free is a free multimodal model designed for image perception and reasoning. Free model availability changes over time, so check OpenRouter's free models for current alternatives. Free endpoints use shared capacity and may respond more slowly or time out under load. When using this server, set imageAnalysisTool to openrouter_image_analyze_image in the plugin config.

Note

Any MCP server with an image analysis tool will work. The servers above are examples. For a different tool, set imageAnalysisTool; see Configuration.

Note

OpenCode v2 hides MCP tools behind Code Mode by default, so the model can only reach them by writing code for an execute tool. The plugin turns Code Mode off for the server that provides the vision tool, so the model can call it directly. Other servers are left alone, and a codemode value you set on that server always wins. See disableCodeMode.

2. Install the plugin

Install it for all projects with the OpenCode CLI:

opencode plugin add opencode-easy-vision@next

Or add it to the plugins array in your opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugins": ["opencode-easy-vision@next"]
}

To work on a local checkout, see CONTRIBUTING.md.

---

Agent-Assisted Setup

Paste this into OpenCode and let your agent handle the rest:

Set up opencode-easy-vision by following https://raw.githubusercontent.com/devadathanmb/opencode-easy-vision/main/AGENT_SETUP.md

What the agent will do

The agent follows the instructions in AGENT_SETUP.md. It will:

1. Inspect your existing OpenCode and plugin configuration
2. Ask whether to set up the plugin for all projects or only the current project
3. Ask which MCP image analysis tool and models you want to use
4. Update existing files in place instead of replacing them
5. Summarize the changes and guide you through verification after a restart

Usage

1. Select a configured model in OpenCode.
2. Paste an image (Cmd+V / Ctrl+V).
3. Ask your question, just like you would with Claude or GPT.

Configuration

Important

By default, this plugin only activates for MiniMax provider models, where MiniMax is the direct provider in OpenCode. These IDs match minimax/*, minimax-cn/*, and related patterns. If you access a MiniMax model through a third-party provider such as OpenRouter, or use another model, add that model's pattern to models. See CONFIGURATION.md.

Each option uses the following priority order:

1. Project level: .opencode/opencode-easy-vision.jsonc
2. User level: ~/.config/opencode/opencode-easy-vision.jsonc

On first load, an example config is created at ~/.config/opencode/opencode-easy-vision.jsonc. See CONFIGURATION.md for config precedence and the full option reference.

Supported Image Formats

PNG, JPEG, and WebP. Exact format support depends on the image analysis tool you configure.

Troubleshooting

Plugin not activating?

By default, the plugin only activates for MiniMax provider models, with IDs matching minimax/*, minimax-cn/*, and related patterns. It does not activate for a MiniMax model accessed through OpenRouter or another provider. To use the plugin with a different model or provider, add the model's ID pattern to models in your config. See CONFIGURATION.md.

Model says the vision tool is unavailable?

OpenCode names MCP tools <server>_<tool>, so the MiniMax server above exposes MiniMax_understand_image. Make sure imageAnalysisTool matches that name exactly (it is case-sensitive) and that the MCP server is connected.

Plugin not updating after a new release?

Run opencode plugin update. OpenCode v2 runs plugins in a background service, so restarting the terminal is not enough: run opencode service restart to load the new version.

Using a remote or containerized MCP server?

Images are saved on the machine running OpenCode. The MCP server must be able to read files at those paths.

Where are the logs?

In OpenCode's log file (~/.local/share/opencode/log/opencode.log by default), on lines tagged plugin=easy-vision.

Upgrading from 1.x

1. Upgrade to OpenCode v2, then change your plugin entry to opencode-easy-vision@next. The old "plugin" key still works, though "plugins" is the v2 key.
2. OpenCode v2 names MCP tools differently. If your config sets imageAnalysisTool to mcp_minimax_understand_image, remove it to use the new default, MiniMax_understand_image, or set it to your server's <server>_<tool> name.
3. Config files must now be named opencode-easy-vision.jsonc. Rename any opencode-easy-vision.json or opencode-minimax-easy-vision.json(c) files; their contents do not need to change.

Uninstallation

1. Run opencode plugin remove opencode-easy-vision, or remove it from the plugins array in your opencode.json.

2. Delete the plugin config files you no longer need:

rm -f ~/.config/opencode/opencode-easy-vision.jsonc
rm -f .opencode/opencode-easy-vision.jsonc

Saved images live in opencode-easy-vision/ in your OS temp directory (or your configured tempDir) and are cleaned up automatically.

Contributing

See CONTRIBUTING.md for local development setup.

License

AGPL-3.0. See LICENSE.

References

• OpenCode Official Website
• OpenCode V2 Plugins Documentation
• MiniMax Coding Plan MCP Repository
• openrouter-image-mcp Repository