# Vision Model Text Input Generator MCP

An MCP server that renders text into PNG canvases for local vision models. It is based on the output behavior of the adjacent `Vision-Model-Text-Input-Generator` app.

## Hosted MCP endpoint

The production Streamable HTTP endpoint is deployed separately from LibreChat:

```text
https://vision-model-text-input-generator-mcp-1080561124190.us-central1.run.app/mcp
```

It requires `Authorization: Bearer <MCP_AUTH_TOKEN>`.

## Local development

```bash
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
MCP_AUTH_TOKEN=local-development-token python main.py
```

The server exposes `generate_vision_text_image`. Its image result is standards-compliant MCP `ImageContent` (`image/png`), so capable clients such as LibreChat can render it directly.

## LibreChat configuration

Add the following under `mcpServers` in `librechat.yaml`:

```yaml
mcpServers:
  vision-text-image:
    title: Text to Image Text
    type: streamable-http
    url: https://vision-model-text-input-generator-mcp-1080561124190.us-central1.run.app/mcp
    requiresOAuth: false
    headers:
      Authorization: 'Bearer ${VISION_TEXT_IMAGE_MCP_TOKEN}'
    serverInstructions: true
```

Restart LibreChat, then add **vision-text-image** to an agent's MCP tools. Use a vision-capable chat model when you want the generated image to be included in the model's immediate context.

## Verify

Run the supplied local checks after installing dependencies:

```bash
python -m py_compile main.py
python verify_mcp.py
```