[eLyiN]

Root cause

Third-party directories and the README tell users to run uvx codex-bridge. That install
resolved the unbounded mcp>=1.0.0 dependency to the 2.x SDK, where FastMCP no longer
exists, so the published package died at startup:

ModuleNotFoundError: No module named 'mcp.server.fastmcp'. This is mcp 2.x, where FastMCP
was renamed to MCPServer (from mcp.server.mcpserver import MCPServer) ... or pin 'mcp<2' to
keep running v1 code.

Fix

• Migrated src/mcp_server.py to the 2.x SDK: from mcp.server.mcpserver import MCPServer,
  mcp = MCPServer("codex-assistant").
• Bounded the dependency to mcp>=2,<3 in pyproject.toml, .github/workflows/ci.yml
  (which installed mcp>=1.0.0 literally) and .github/workflows/release.yml.
• Added tests/test_mcp_handshake.py: an offline, in-process MCP initialize handshake that
  lists the tools, so this class of breakage fails CI instead of shipping.
• Version bumped to 1.2.4 in pyproject.toml and src/__init__.py, with a CHANGELOG entry.

Verification (real output, not claims)

• Clean venv, pip install . on the branch, resolving mcp 2.2.0: console script completes the
  handshake and returns serverInfo.name = codex-assistant with the three tools
  (consult_codex, consult_codex_with_stdin, consult_codex_batch).
• python tests/test_mcp_handshake.py passes in 0.01s.
• Before the fix, the same console script on the published 1.2.3 exited 1 with the import error
  above.

Not in this PR

The PyPI release (tag v1.2.4) is deliberately not pushed. That step needs explicit human
approval; after merge, git tag v1.2.4 && git push origin v1.2.4 triggers the release workflow.