memoir-mcp provides persistent memory for AI agents across sessions by logging attempts, blockers, decisions, and handoffs to a local SQLite database — no external services or API keys required.

Write Tools:

• log_attempt — Record something that was tried along with its outcome

• log_blocker — Flag an obstacle and document why it's blocked

• resolve_blocker — Mark a blocker as resolved by providing its ID and what fixed it

• log_decision — Record a design or architectural decision and its rationale

• end_session — Explicitly close the current session with an optional summary

Read Tools:

• get_handoff — Retrieve a structured summary of the last session (attempts, blockers, decisions) to quickly get up to speed

• get_history — Query past sessions for the current project (default: last 3, up to 20)

• get_blockers — List unresolved or resolved blockers across all sessions for the project

Key behaviors:

• Auto-detects projects via git root and implicitly creates sessions

• Stores all data locally in ~/.memoir/memoir.db

• Retains up to 20 sessions per project, automatically pruning older ones

• Configurable via environment variables (MEMOIR_MAX_CONTENT, MEMOIR_MAX_OUTCOME, MEMOIR_MAX_ENTRIES, MEMOIR_MAX_SESSIONS)

• Compatible with any MCP client (Claude Code, Cursor, Codex, Windsurf, etc.)

Uses Git repository roots to provide automatic project detection, ensuring that session journals and persistent memory logs are correctly scoped to the current project.