Memharness
- 4 repo stars
- Updated June 26, 2026
- las7/memharness
Memharness is a bi-temporal, provenance-carrying memory system for AI agents that stores facts in one SQLite file so clients can query beliefs, changes, and sources over time via MCP.
About
Memharness is an MCP-backed memory primitive for AI agents that stores facts in a single SQLite database with bi-temporal semantics and full provenance. It answers audit-style questions—what was believed at a date, what changed, and why—without LLM calls in the storage layer. Use it when agent history outgrows the context window or you need explainable, correctable long-term memory. Skip it for small, static context that fits comfortably in CLAUDE.md or the prompt.
- Bi-temporal facts: valid time vs learned time
- Supersession links corrections without silent deletes
- Per-fact provenance and session-scoped forget
- SQLite-only storage layer; no LLM on write path
- MCP server exposes seven memory tools
Memharness by the numbers
- Data as of Jul 7, 2026 (Skillselion catalog sync)
claude mcp add --env MEMHARNESS_DB=YOUR_MEMHARNESS_DB memharness -- npx -y @memharness/mcpAdd your badge
Show developers this MCP server is listed on Skillselion. Paste this into your README.
| repo stars | ★ 4 |
|---|---|
| Package | @memharness/mcp |
| Transport | STDIO |
| Auth | None |
| Last updated | June 26, 2026 |
| Repository | las7/memharness ↗ |
How do you give agents reliable long-term memory with an audit trail when string bags and plain project files cannot answer as-of or why queries?
Give agents durable, auditable long-term memory in one SQLite file with bi-temporal recall and provenance via MCP.
Who is it for?
Engineers building agents that accumulate months of facts, need temporal recall, diff, and provenance-scoped deletion, and want self-hosted deterministic storage.
Skip if: Short-lived agents with tiny static memory where stuffing context or a single markdown file is simpler and sufficient.
What you get
A persisted SQLite memory store with MCP tools for recording facts, ranked recall, as_of snapshots, diffs, why explanations, and provenance-filtered retraction.
Recommended MCP Servers
0Latency MemoryPersistent memory layer for AI agents.
0nMCP — Universal AI API OrchestratorUniversal AI API Orchestrator — 1,554 tools, 96 services.
0xHumans Protocol MCPMCP for AI agents: financing, skills, lending on Base1k Patient Mcp1k patient MCP server
1trippulse1trip PULSE: 21-tool AI travel planner.
3D AI Agent Avatar3D AI Agent Avatar — render any GLB, give it a Solana wallet, a voice, and pump.fun powers.89FAQ
Does memharness call an LLM when writing memories?
No—the storage layer is deterministic; the client or skill decides what facts to write without model extraction in core.
How is this different from mem0 or Zep?
It focuses on temporal and provenance queries in embeddable SQLite rather than hosted LLM-driven extraction pipelines.
What do I install to use it with Claude or other MCP clients?
The @memharness/mcp package exposes stdio MCP tools on top of the core library schema and write path.