
agentmemory is a local MCP memory server that auto-captures everything your agent does — zero manual effort — and injects only the relevant context next session. Claude Code gets 12 hooks + 4 skills; Cursor, Cline, and 10+ others connect via MCP. At 240 observations, it uses ~1,900 tokens vs. 22,000+ for CLAUDE.md. Hit #1 GitHub Trending with 11K+ stars.
.cursorrules, Cursor notepads — caps out around 200 lines and dumps everything into context whether it's relevant or not. At 240 saved observations, that's 22,000+ tokens per session just for context overhead.jose over jsonwebtoken for Edge compatibility, or that tests live in test/auth.test.ts. You spend the first five minutes re-explaining context.npx @agentmemory/agentmemoryhttp://localhost:3113 to watch memory build in real time.npx @agentmemory/agentmemory demoInstall agentmemory: run `npx @agentmemory/agentmemory` in a separate terminal to start the memory server. Then run `/plugin marketplace add rohitg00/agentmemory` and `/plugin install agentmemory` to register all 12 hooks, 4 skills, and 43 MCP tools. Verify with `curl http://localhost:3111/agentmemory/health`.{
"mcpServers": {
"agentmemory": {
"command": "npx",
"args": ["-y", "agentmemory-mcp"]
}
}
}gemini mcp add agentmemory -- npx -y agentmemory-mcpnpm install @xenova/transformersiii-engine runtime. Download iii-x86_64-pc-windows-msvc.zip from the iii-hq/iii releases page, extract iii.exe to your PATH, then run the npx command above. If you'd rather skip the binary, Docker Desktop works as a fallback — agentmemory auto-detects it.1| Skill | What it does |
|---|---|
/recall jwt auth | Semantic search across all past sessions — finds "N+1 query fix" when you type "database performance" |
/remember "using jose over jsonwebtoken for Edge compatibility" | Saves a durable fact to long-term memory |
/session-history | Summarizes recent sessions so you can orient quickly |
/forget | Deletes specific observations or full sessions |
memory_recall, memory_save, memory_smart_search, and memory_file_history tools — 43 MCP tools total, though the default install exposes only the 7 most useful ones. Set AGENTMEMORY_TOOLS=all in ~/.agentmemory/.env to unlock the full set.1| agentmemory | Built-in (CLAUDE.md) | mem0 (53K ★) | |
|---|---|---|---|
| Retrieval R@5 | 95.2% | N/A (loads all) | 68.5% |
| Tokens at 240 observations | ~1,900 | 22,000+ | Varies |
| Auto-capture | 12 hooks, zero effort | Manual editing | Manual add() calls |
| Cross-agent | MCP + REST | Per-agent files | API only |
| Self-hosted | Yes (default) | Yes | Optional |
| External dependencies | None (SQLite + iii-engine) | None | Qdrant / pgvector |
"The developers who figure out how to build great agent-native workflows — not just prompts, but real systems — are going to have a serious competitive advantage in the next 2–3 years."3
"Remembering everything can be as bad as forgetting everything. [...] It started over-weighting old objections, treating one enterprise prospect's complaint like a market trend."5
/forget to prune low-signal noise.@agentmemory/agentmemorygithub.com
producthunt.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.