Patch release focused on exposing Emotion Engine through a local stdio MCP server and making client registration harder to misconfigure.
Highlights
- Added a zero-dependency local stdio MCP server for Emotion Engine runtime/protocol tools, including status, compact summaries, record policy, advisory appraisal, session lifecycle, turn recording, trust settlement, and recent log inspection.
- Bundled the MCP server into the Codex skill package while keeping Agent Harness responsibilities out of the MCP boundary: install refresh, doctor, repair, manifest checks, and sidecar drift remain Agent Harness work.
- Added MCP client registration examples for Codex CLI, Claude Code, Claude Desktop, and project-level
.mcp.json. - Added
scripts/register_mcp_client.pyso users can register the local stdio server without hand-writing MCP config. - Clarified state path rules: generic projects use
.emotion-engine/emotion-state.json, while Codex and Agent Harness targets use.emotion-engine/codex-state.jsonso wrappers and MCP clients do not split state.
Verification
PYTHONDONTWRITEBYTECODE=1 python3 tests/test_mcp_registration.py-> 6 tests OKPYTHONDONTWRITEBYTECODE=1 python3 tests/test_mcp_server.py-> 5 tests OKPYTHONDONTWRITEBYTECODE=1 python3 tests/test_demo_static.py-> 6 tests OKpython3 -m py_compile scripts/emotion_engine_mcp.py scripts/register_mcp_client.py tests/test_mcp_server.py tests/test_mcp_registration.py-> OKPYTHONDONTWRITEBYTECODE=1 python3 tests/test_codex_integration.py-> 7 tests OKPYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover -s tests-> 82 tests OKgit diff --check-> OK