v0.4.0 - MCP Protocol Support

What's Changed

• Rewrite server.mjs to support MCP stdio protocol (JSON-RPC over stdio) as primary mode
• Keep HTTP mode available via --http flag for testing
• Support initialize, tools/list, tools/call MCP methods
• Extract tool handling into shared handleToolCall() function
• Bump version to 0.4.0

MCP Protocol

The server now speaks the MCP protocol over stdin/stdout, making it compatible with Codex and other MCP clients. The HTTP API is still available by passing --http flag.

v0.3.1 — README Fix & Documentation Overhaul

What's Changed

📖 Documentation

• Fixed emoji corruption: All emoji characters in README were corrupted (showing as ???), now replaced with correct Unicode emoji
• Added "What It Does" section: Clear explanation of the proxy/forwarding architecture
• Added parameter table: Detailed image_generate parameters with required/default/description columns
• Added response example: image_config_status now shows example JSON response
• Clarified provider modes: Explicit behavior differences between generic and openai modes
• Updated file structure: Accurate descriptions for all files in the project

🔗 Links

• Full Changelog

v0.3.0 — Test Fix & Version Bump

What's Changed

🐛 Bug Fixes

• test_server.py: Auto-set IMAGE_GEN_BASE_URL and IMAGE_GEN_API_KEY environment variables before running, so smoke tests pass without manual env setup

📦 Version

• Bumped version to 0.3.0 in plugin.json and README badge

🧪 Verified

• Plugin validation: VALIDATION OK
• Smoke test: All 4 checks pass (health, config_status, missing prompt, upstream error)
• Provider mode tests: Both openai and generic modes pass with correct payload normalization

Image Generation MCP v0.2.0

Pure image-generation MCP only. Uses IMAGE_GEN_BASE_URL and IMAGE_GEN_API_KEY to call a configured image API and exposes image_generate plus image_config_status. No planner/subagent workflow.