The server provides persistent, searchable long-term memory for AI coding agents, running locally with an embedded SQLite database (no external database or API key required).

• Memory Operations: Recall past memory using hybrid BM25 + vector search with pagination/token limits; capture decisions, learnings, tasks, errors, conversations, and atomic facts; search by keyword/semantic similarity with filters; forget by ID, tag, type, or date with confirmation.

• Agent Handoff & ADRs: Create structured handoff packets to transfer context between sessions/agents (saving 60–85% tokens); record architecture decision records with rationale, alternatives, and consequences.

• Data Management: Export/import memory as JSON (filterable, stdout support); backup database and audit log; view memory in a local web viewer at http://localhost:7331; print stats (counts, top tags, sessions, etc.).

• Automation: Install agent skills and lifecycle hooks (SessionStart auto-recall, Stop handoff reminder) for Claude Code, Devin CLI, and others.

• Optional LLM Pipeline: With an API key, distill raw captures into atomic facts (L1), scenarios (L2), and persona profiles (L3); otherwise, raw captures and hybrid search work in noop mode.

• Security: Automatic secret redaction on captures; token caps on recall (4,000) and search (8,000); audit log of all tool calls with hashed arguments.

• Team Collaboration: Sync shared team memory via a committed .tdai-memory/memory-export.json file in a repo.

• Programmatic Access: TypeScript SDK for direct capture, recall, handoff, adr calls.

• Configuration: Configurable via environment variables or config file; supports multiple clients (Claude Code, Cursor, Codex CLI, Devin CLI, Trae); Docker support.

Syncs project memory to a Git repository, enabling team sharing through commits, diffs, and merges.

1. Click on "Install Server".

2. Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.

3. In the chat, type @ followed by the MCP server name and your instructions, e.g., "@tdai-memory-mcp what did we decide about the API design?"

That's it! The server will respond to your query, and you can continue using it as needed.

Here is a step-by-step guide with screenshots.