feat: v1.1.2 - MCP file context support

- Add 'files' parameter to quorum_discuss for passing file paths
- MCP server reads files and includes them as context
- Limits: max 10 files, 100KB per file, 500KB total
- Auto GitHub Releases on tag push

fix: v1.1.1 - MCP error handling and test fixes

- Add try/except for MCP runtime errors (API failures, timeouts)
- Fix _write_json to handle StringIO in tests (stdout.buffer fallback)
- Exclude MCP module from pytest collection (missing mcp dependency)
- Improve quorum_discuss tool description with model requirements

release: v1.1.0 - MCP server integration

Add Model Context Protocol (MCP) server for Claude Code/Desktop integration:
- New `quorum-mcp-server` entry point
- Tools: quorum_discuss, quorum_list_models
- Resources: quorum://models, quorum://methods
- Compact output mode (synthesis only) by default
- Full output available via full_output parameter

Usage: claude mcp add quorum -- quorum-mcp-server

release: v1.0.7 - PyPI distribution with first-run setup

Added:
- PyPI distribution: pip install quorum-cli && quorum
- First-run setup: creates ~/.quorum/.env.example automatically
- Config loaded from ~/.quorum/.env (works from any directory)
- GitHub Actions workflow for automated PyPI publishing

Fixed:
- Startup spinner shows immediately on all platforms
- /help and /status modals close with ESC
- Build wheel directly (not from sdist) to include frontend

release: v1.0.6 - PyPI distribution support

Added:
- PyPI distribution: `pip install quorum-cli && quorum` works out of the box
- GitHub Actions workflow for automated PyPI publishing (Trusted Publishing)
- Frontend bundled into wheel automatically

Fixed:
- Startup spinner now shows immediately on all platforms
- /help and /status modals can be closed with ESC key

Changed:
- Simplified launcher scripts with signal file coordination
- Frontend signals readiness for seamless spinner-to-UI transition

fix: Windows native support (v1.0.5)

- Fixed PROJECT_ROOT detection using import.meta.url instead of process.cwd()
- Fixed Python asyncio stdin reading using run_in_executor (cross-platform)
- Python's connect_read_pipe doesn't work on Windows:
  - ProactorEventLoop: WinError 6 with pipe handles
  - SelectorEventLoop: connect_read_pipe not implemented
- Quorum now works correctly when launched from quorum.bat on Windows

feat: filter non-generative Ollama models (v1.0.4)

Embedding and whisper models are now automatically filtered from
the Ollama model discovery. These models cannot participate in
discussions as they don't generate text responses.

Filtered patterns:
- embed (nomic-embed-text, mxbai-embed-large, etc.)
- bge- (bge-m3, bge-large)
- minilm (all-minilm)
- paraphrase (paraphrase-multilingual)
- whisper (speech-to-text models)

Vision models (llava, gemma3, etc.) are NOT filtered - they can
generate text.

feat: adaptive execution mode + OpenAI-compatible providers (v1.0.3)

Adaptive Execution Mode:
- Smart VRAM management for local Ollama models
- QUORUM_EXECUTION_MODE: auto, parallel, sequential
- Auto mode runs cloud APIs parallel, Ollama sequential

OpenAI-Compatible Providers:
- OpenRouter: 200+ models via one API key
- LM Studio: Local models with GUI
- llama-swap: Hot-swap server for local models
- Custom: Generic OpenAI-compatible endpoint
- Smarter display name formatting for namespaced models

release: v1.0.1 - JSON export with method-specific schemas

Add structured JSON export with dedicated schemas for each discussion method:
- 7 method-specific parsers (Standard, Oxford, Advocate, Socratic, Delphi, Brainstorm, Tradeoff)
- 9 schema files defining export document structure per method
- Synthesis/verdict/decision now included in final phase messages (no empty phases)
- Unified export architecture: PDF and JSON use same method-aware parser
- Remove legacy generic parser

v1.0.0 - Initial release