Dense-MemMCP Server

Dense-Mem is a durable memory layer for AI agents that treats memory as something to audit, not just retrieve. Rather than letting an agent's memory silently drift as new, possibly wrong, statements come in, it stores each piece of information as evidence, tracks provenance back to the source, and explicitly flags when new input contradicts what is already known instead of overwriting it.

Over MCP it exposes tools such as remember to store new evidence and report where it landed, recall_memory to retrieve relevant facts and claims, confirm_memory to resolve a flagged contradiction with a user decision, trace_memory to walk a fact back through its evidence chain, reflect_memories to surface stale or low-confidence items for review, and import/export tools for moving a memory pack between environments or teammates. It is built on PostgreSQL with pgvector for embedding-based recall, and ships as a Docker Compose stack with its own lightweight web UI for browsing and correcting memory alongside the MCP surface. This targets teams building longer-running or multi-session agents, such as support bots, research assistants, and coding agents, that need memory to survive restarts and stay trustworthy rather than accumulating silent errors. Embeddings can come from OpenAI-compatible APIs or a local Ollama instance, so it can run fully offline.