360 Secure SkillHub
360 安全技能中心

lodekeep

该 Skill 通过远程 Lodekeep MCP 服务提供跨会话记忆：可检索历史决策、捕获带提交溯源的新笔记、加载项目上下文与记忆地图，并管理项目级人格。需 Lodekeep 账号与 API Key，笔记写入远程服务；不包含本地存储或自动捕获机制。

steve0no
steve0no
热度 15.33
最近更新 2026-09-29 20:04:15
分类：知识依赖：MCP 服务 / 第三方 API应用场景：笔记整理 / 研究 / 工作流

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add lodekeep
Metadata
name
lodekeep
version
1.0.0
description
Persistent cross-session memory for OpenClaw agents via Lodekeep MCP. Your agent survives compaction, session death, and tool switches with its decisions intact - every claim carries the exact commit it was verified against.
homepage
https://lodekeep.com
metadata
{"openclaw":{"emoji":"🦞","category":"memory"}}

Lodekeep — one memory for every AI

Your agent's decisions die in three places: compaction eats them, session death orphans them, and tool switches (Claude Code → Gemini CLI → another agent) strand them in incompatible histories. Lodekeep is a durable memory layer over MCP that all of your tools read and write, so a decision captured once shows up everywhere.

Why this exists (the receipts doctrine)

Memory that only says "we decided X" drifts. Lodekeep notes carry provenance: what was decided, why, and the exact revision/artifact it was verified against. When the tree changes, stale notes say so instead of lying. A memory layer is only trustworthy if it distinguishes what was true then from what is true now.

Requirements

  • A Lodekeep account (free tier works): sign up at https://lodekeep.com
  • An MCP-capable client: OpenClaw gateway, Claude Code, Gemini CLI, or any MCP client
  • Your Lodekeep API key (dashboard → Settings → API keys)

Setup (two minutes)

  1. Create an account at https://lodekeep.com and copy your API key from Settings.

  2. Add the Lodekeep MCP server to your agent's MCP config (it's a remote server — no npm install needed):

{
  "mcpServers": {
    "lodekeep": {
      "url": "https://lodekeep.com/mcp",
      "headers": {
        "Authorization": "Bearer your-key-here"
      }
    }
  }
}

For OpenClaw specifically, add the same block to the gateway's MCP server config (~/.openclaw/openclaw.json → mcpServers), then restart the gateway.

  1. Verify: ask your agent to "check lodekeep server_info" — it should report authenticated with your project list.

API Key Discovery

Check in this order (never hardcode):

  1. Environment variable: LODEKEEP_API_KEY or MCP_LODEKEEP_API_KEY
  2. Config file: ~/.config/lodekeep/credentials.json or ~/.openclaw/openclaw.json env block
  3. If no key is found, ask the user to provide one before proceeding.

What your agent can do once connected

Tool Use it to
search_memory Recall any past decision by meaning — "what did we decide about rate limits?"
capture_note Persist a decision/pattern/lesson mid-task so it survives compaction
get_project_context Load a project's curated overview at session start
get_memory_map See what memory exists before duplicating work
list_personas / set_project_persona Persistent operating personas

Operating rules (the part that makes it work)

  1. Capture on decisions, not chatter. A note is for decisions, corrections, and lessons — things that change future behavior.
  2. Search before you write. If a note exists, update it instead of duplicating.
  3. Provenance always. A note without "verified against X at commit Y" is a claim, not a receipt.
  4. Route notes to a project ("Steve - Repo Sentinel Monetization", "lodekeep-launch") so recall stays precise.

Known limitations (honest ones)

  • Notes are semantic-recalled, not exhaustively indexed — very large projects should keep curated projects, not firehoses.
  • Live secrets never belong in memory. Store where the credential lives, not its value.
  • Per-project recall requires the project to be marked included in your dashboard.

Links