NEW · MCP support for Claude Code, Cursor, and OpenClaw. Every context, connected. Every agent, empowered.
This repository is the official skill / rules / instructions distribution for Contextberg — drop it into your project and your AI coding agent will know how to read your activity history, daily reports, and Long-Term Memory through the Contextberg MCP server.
Contextberg is a local memory app for your AI agents. It watches your screens, agent transcripts, and browser in the background — so Claude Code, Cursor, OpenClaw, and other agents can just remember.
| Pillar | What it does |
|---|---|
| 01 · Context | Records screens, inputs, browser activity, and agent transcripts — delivered to your agent via MCP. No config files, just connect. |
| 02 · Memory | Distills your daily work into three layers — activity (raw timeline), daily (per-day digest), and long-term (your tools and patterns). |
| 03 · Local | Runs entirely on your machine. Pair with LM Studio for a fully offline pipeline — no data ever leaves your PC. |
| 04 · Remember | Auto-summarizes what you were doing before you stepped away, so you pick up right where you left off. Then dig deeper in chat. |
The Contextberg desktop app exposes its memory via a local MCP server at http://localhost:18080. Agents can talk to it directly, but they work much better when they know which tool to call when — get_activity first, then drill into get_daily_memory, get_weekly_memory, get_agent_history, read_ltm, or update_ltm.
This repo packages that guidance as a skill in every popular agent format, so any of these agents can use Contextberg with a single drop-in. Two layouts are provided side by side:
1. Unified <tool>/skills/contextberg/SKILL.md layout (drop-in for tools that follow the new convention):
| Agent | Path |
|---|---|
Generic / .agent aware |
.agent/skills/contextberg/SKILL.md |
Generic / .agents aware |
.agents/skills/contextberg/SKILL.md |
| Claude Code | .claude/skills/contextberg/SKILL.md |
| Cline | .cline/skills/contextberg/SKILL.md |
| Codex | .codex/skills/contextberg/SKILL.md |
| Cursor | .cursor/skills/contextberg/SKILL.md |
| Gemini CLI | .gemini/skills/contextberg/SKILL.md |
| GitHub (Copilot / Models) | .github/skills/contextberg/SKILL.md |
| OpenCode | .opencode/skills/contextberg/SKILL.md |
2. Native per-tool rule formats (for tools that read their own rule file shape):
| Agent | Path |
|---|---|
| Cursor (MDC rules) | .cursor/rules/contextberg.mdc |
| GitHub Copilot (instructions) | .github/copilot-instructions.md |
| Windsurf (rules) | .windsurf/rules/contextberg.md |
| Cline (legacy rules) | .clinerules/contextberg.md |
| Continue (rules) | .continue/rules/contextberg.md |
| OpenClaw / Aider / any AGENTS.md-aware agent | AGENTS.md |
All formats carry the same canonical content (kept in sync) — pick the one(s) your agent reads.
Install this skill straight from GitHub into your current project:
npx skills add contextberg/contextberg-skillsThis drops .claude/skills/contextberg/SKILL.md into your project. Restart Claude Code and the skill is live.
For Cursor, Cline, Codex, Gemini, OpenCode, Windsurf, Continue, Copilot, or any other agent, copy the relevant file or directory from this repo (see tables above) into your project root.
git clone https://github.com/contextberg/contextberg-skills .contextberg-skills
# then symlink or copy the formats you want into your projectThe skill assumes the Contextberg desktop app is running locally. Add the MCP server to your agent's config. Example (Claude Code, .claude/settings.local.json):
{
"mcpServers": {
"contextberg": {
"command": "cmd",
"args": ["/c", "npx", "-y", "@contextberg/mcp-server"]
}
}
}For Cursor, OpenClaw, and other MCP-aware agents, point them at the same npx @contextberg/mcp-server command — exact location depends on the agent.
get_activity ← always start here (full daily picture, includes agent sessions)
├── get_daily_memory → AI-written daily summary
├── get_weekly_memory → multi-day patterns / trends
├── get_agent_history → all turns of one Claude Code / Cursor / OpenClaw session
└── read_ltm / update_ltm → who the user is — preferred stack, work style, current priorities
get_activity returns a single consolidated view: app timeline, keystrokes per region (editor / terminal / AI sidebar), clipboard history, browser visits, and agent session summaries. Most "what was I doing?" questions can be answered from this one call — the rest of the tools exist for AI summaries, multi-day trends, full-turn deep dives, and the persistent user profile.
Open an issue or PR if your agent uses a format that isn't listed here. Keep all formats in sync — the canonical text lives in AGENTS.md; changes there should be mirrored to the per-agent files.
MIT