Claude Code — the thread in your terminal

Claude Code natively reads CLAUDE.md and the project MCP config (.mcp.json). MEMIA projects both from the OPERA workspace: Claude starts every session with the same invariants as all your other surfaces, and can read the memory over MCP.

Install

cd ~/MEMIA
memia projections write     # writes CLAUDE.md, AGENTS.md, .cursor/rules/memia.mdc, .mcp.json
memia projections check     # invariants IDENTICAL ✅

claude                      # launch Claude Code INSIDE the workspace: it reads CLAUDE.md
                            # and offers the « memia » MCP server declared in .mcp.json

What Claude Code can do

• Read the memory over MCP: memia_search, memia_list, memia_read, memia_engagements_digest (read-only).
• Act through the CLI: memia scan, memia engagements confirmer… — every gesture stays journaled.
• Serve as an execution door for harness agents: memia agents run core-search-navigator --prompt "…" --porte claude.

The claude door runs with explicit permissions (allowed tools listed, network tools off, never a bypass): a reader agent cannot write, a writer agent only writes to the OPERA families in its definition.