Power your agent
คำสั่งของผู้เผยแพร่เอง ไม่ใช่เอนด์พอยต์ของ PlugMCP
Host a copy with our keys and logs.
npx -y [email protected]Claude Code ไม่มีลิงก์ติดตั้งสำหรับคำสั่งแบบโลคัล ให้คัดลอกคำสั่งด้านล่างแล้วรันในเทอร์มินัล
claude mcp add mcp-agents-memory -- npx -y mcp-agents-memory@0.9.35รันในเทอร์มินัล
{ "mcpServers": { "mcp-agents-memory": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}claude_desktop_config.json
URL แบบรีโมตหรือคำสั่งบนเครื่องของคุณ? การติดตั้งสองแบบต่างกันอย่างไร
รันบน PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
เปิดจากผู้เผยแพร่
Long-term, time-ordered memory for AI agents — a shared memory pool that persists across sessions and machines, modeled on human memory.
Multiple agents (Claude Code, Codex, Gemini CLI, Grok, Antigravity, …) share one memory pool, accumulating and recalling memories in chronological order. Each platform/model gets its own view automatically, while project tags let collaborators see each other's relevant context.
Long-term memory for AI agents with provenance tracking
🇰🇷 한국어 README →
README.ko.mdDesign rationale and decisions →RESPEC.md
MEMORY.md-style self-updating memory, skills & rulesThis project aims to combine the strengths of both: supermemory's recall and Hermes' self-curation, without their weaknesses (machine-locked storage, opaque mutation).
fact_type taxonomy. Older memories are summarized by tag and archived (soft-delete, never destroyed).agent_platform / agent_model columns alone separate memories per model. No manual categories.┌─────────────────┐ ┌──────────────────────────┐
│ Agent │ │ MCP Server │
│ (Claude Code, │ ───▶ │ ▶ Hot Path (instant save)│ ──▶ memory table
│ Codex, ...) │ └──────────────────────────┘ (raw + role + platform/model)
└─────────────────┘ │
│ rows with NULL p_tag/d_tag/embedding accumulate
▼
┌──────────────────────────────┐
│ Cold Path (1 min / 5 msgs) │
│ ├─ Tagger → p_tag, d_tag │
│ ├─ Embedder → embedding │
│ ├─ Librarian → user profile │
│ ├─ Clusterer → tag summaries │
│ └─ AliasPromoter → tag merges │
└──────────────────────────────┘
The Cold Path's LLM roles (tagger / librarian / clusterer / project-alias judge) all run on a single shared backend — local Qwen3-14B via llama.cpp, or a cloud fallback (see Cold Path LLM backend).
memory — raw, time-ordered conversation log (single table, soft-delete archive)
users — core facts the Librarian promotes out of memory
project_tags — project tags, grown dynamically by the Cold Path
When several machines share one database, the Cold Path (tagging / profiling / clustering / alias judging) must run on exactly one of them — otherwise machines double-process the same rows and double the cloud cost. The same package splits roles purely by config:
COLD_PATH_ENABLED=false.mcp-agents-memory coldpath # Cold Path worker only, no MCP server (systemd recommended)
mcp-agents-memory buzz-ingest copies chat messages from the Buzz rooms a given identity is a member of into memory, one row per message, through the official buzz CLI. It also strips the quoted history from captured Buzz turns once every quoted message exists as a row (the original stays in raw_message; --no-reclean turns this off). Run it from a timer; see CHANGELOG 0.9.20–0.9.21 for the settings.Point LOCAL_LLM_BASE_URL at any OpenAI-compatible endpoint to use local / self-hosted inference (llama.cpp, ollama, …). If unset, a cloud model is used; LOCAL_GROK_FALLBACK=true falls back to Grok when local inference fails.
e.g. serve
Qwen3-14Bwith llama.cppllama-serveron an AMD/NVIDIA GPU and setLOCAL_LLM_BASE_URL=http://localhost:8080/v1→ Cold Path cloud cost ≈ $0. (Ajson_schemagrammar with thinking disabled guarantees valid JSON.)
chars / 1.7) to protect Hot Path latency. Window is env-tunable.agent_platform / agent_model. A p_tag match (same project) pulls in collaborating agents' memories too.ILIKE (env-tunable, starts at 0.3).The server exposes 4 tools. Most clients only ever need the first three; save_message is a fallback.
memory_startup — session boot briefReturns a markdown brief (recent conversations, active projects, user profile) so a new session picks up where the last left off. On supported clients it is injected automatically at connect; call it explicitly to refresh mid-session.
search_memory — unified read / searchsearch_memory({
query?: string, // semantic search (vector + ILIKE fallback)
p_tag?: string, // restrict to a project
date_range?: string, // e.g. "2026-04-29..", "last_week"
role?: 'user' | 'assistant',
agent_platform?: string, // restrict to a platform; omit or '*' = all
device_scope?: 'local' | 'global', // 'global' (default) = all machines; 'local' = this one
limit?: number, // default 10, max 50
include_archived?: boolean,
})
"When you don't remember, reach for this one." Agents just vary the parameters.
manage_knowledge — unified write / editmanage_knowledge({
action: 'add' | 'update' | 'remove',
target: 'sub_profile' | 'memory',
content: string,
})
Use when the user explicitly says "remember this" / "forget that".
target='memory'= force-remember (is_pinned=true, importance bump, archive-exempt).manage_knowledgeskips the Cold Path and syncs tag + embedding immediately, so the memory is searchable the instant you say "got it".
save_message — transcript fallbackFor platforms that don't auto-capture transcripts, the agent calls this each turn to persist the message. On auto-capturing clients (see below) it must not be called — that would duplicate rows.
Auto-injection of the startup brief / auto-capture depends on the client, not the transport — some clients (e.g. desktop/web) don't expose those hooks, so they fall back to explicit tool calls.
See .env.example for the full, annotated list. The essentials:
# DB
DB_HOST=... DB_PORT=5432 DB_USER=... DB_PASS=... DB_NAME=...
# Keys
OPENAI_API_KEY=... # embedding (required)
XAI_API_KEY=... # Grok (Cold Path cloud + local fallback)
# Cold Path LLM backend — OpenAI-compatible endpoint for local inference (omit → cloud)
LOCAL_LLM_BASE_URL=http://localhost:8080/v1
LOCAL_GROK_FALLBACK=true
TAGGER_PROVIDER=local TAGGER_MODEL=qwen3-14b
LIBRARIAN_PROVIDER=local LIBRARIAN_MODEL=qwen3-14b
LIBRARIAN_ENABLED=true
# Hot/Cold path control
COLD_PATH_ENABLED=true # false = client terminal (no Cold Path); only the server is true
COLD_PATH_INTERVAL_SEC=60
COLD_PATH_BATCH_SIZE=5
# Memory load tunables
SHORT_TERM_DAYS=3
SHORT_TERM_TOKEN_LIMIT=8000
SEARCH_FALLBACK_THRESHOLD=0.3
# Agent identity (caller self-reports)
AGENT_PLATFORM=claude-code
AGENT_MODEL=opus-4-8
# ~/.codex/config.toml
[mcp_servers.mcp-agents-memory]
command = "mcp-agents-memory"
args = []
// ~/.gemini/settings.json
{
"mcpServers": {
"mcp-agents-memory": {
"type": "stdio",
"command": "mcp-agents-memory",
"args": [],
"env": {},
"trust": true
}
}
}
Install globally with
npm i -g mcp-agents-memory, or pointcommandat a localbuild/index.js.
Solving the problem in front of you must not break the whole structure.
Every change is checked against the RESPEC.md vision — "does this fix fit the big picture?" — before proceeding. "Just make it run" is a stop signal.
RESPEC.md — current vision, decisions, implementation detail (single source of truth)DEVLOG.md — operational issues, observations, ideasREADME.ko.md — Korean READMELong-term memory for AI agents with provenance tracking
{ "mcpServers": { "mcp-agents-memory": { "command": "npx", "args": [ "-y", "[email protected]" ] } }}เครื่องมือ
ยังไม่มีรายการเครื่องมือ เซิร์ฟเวอร์แบบรีโมตจะได้รายการหลังการตรวจสอบสถานะครั้งแรก
FAQ
คัดลอกคอนฟิก stdio สำหรับ Cursor, Claude Desktop, Claude Code หรือ VS Code ได้ในหน้านี้ คอนฟิกนี้ใช้การติดตั้งของผู้เผยแพร่ ไม่ใช่พร็อกซีของ PlugMCP
ระเบียนนี้ไม่ได้ระบุตัวแปรสภาพแวดล้อมที่จำเป็น
MCP Agents Memory ทำงานแบบโลคัลผ่าน stdio
ตรวจสอบล่าสุดเมื่อ 2026-11-13 Long-term memory for AI agents with provenance tracking