Cairntir

Name: Cairntir
Author: pnmcguire480

@pnmcguire480

关于 Cairntir

Memory-first reasoning layer for Claude Code. Kills cross-chat AI amnesia. MCP server, verbatim storage, prediction-bound drawers, belief-as-distribution retrieval.

#memory #reasoning #verbatim #persistence #local-first #claude-code #knowledge-graph

基本信息

分类

AI 与智能体

许可证

MIT

运行时

python

传输方式

stdio

发布者

pnmcguire480

提交者

Patrick McGuire

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "cairntir": {
      "command": "cairntir-mcp"
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Cairntir?

Cairntir is a local-first, open-source memory system for Claude Code and other Model Context Protocol (MCP) clients. It stores every decision, fact, and lesson that a model writes down — verbatim, with no summarization — in a SQLite database, so that each new chat session inherits the full context from previous ones. It is designed for solo developers, small teams, tool builders, and researchers who want to eliminate the productivity tax of re‑explaining past work across chat sessions.

How to use Cairntir?

Install with pip install cairntir, then run cairntir setup — an interactive wizard that configures the MCP registration in one command. Fully quit and reopen Claude Code (or any MCP client). The server automatically registers a stable launcher and silently re‑registers if the client can’t find it. Six MCP tools are available over stdio: cairntir_session_start, cairntir_recall, cairntir_remember, cairntir_timeline, cairntir_audit, and cairntir_crucible. Two environment variables (CAIRNTIR_DISABLE_AUTOREGISTER and CAIRNTIR_DISABLE_UPDATE_CHECK) allow opt‑out for CI or air‑gapped use.

Key features of Cairntir

Verbatim storage – no summarization, full context preserved forever.
Four‑layer retrieval: identity, essential, on‑demand, and deep.
Semantic search via sqlite-vec for fuzzy recall.
Prediction‑bound drawers with belief‑as‑distribution ranking.
Portable signed format (content‑addressed, HMAC‑signed) for sharing.
Consolidation pass with forgetting curve and contradiction detection.
One‑command installation wizard (cairntir setup).

Use cases of Cairntir

Avoiding the Monday re‑explanation cycle for daily Claude Code users.
Providing a per‑developer memory layer that can later become a shared, gossip‑able memory for small teams.
Embedding persistent context into products that use an MCP client.
Tracking which hypotheses held and which failed with prediction‑bound drawer schema (experiment journal).
Supporting longer‑term projects such as post‑scarcity manufacturing research (horizon goal).

FAQ from Cairntir

What makes Cairntir different from other memory tools?

Cairntir stores everything verbatim, offers a four‑layer retrieval taxonomy, uses semantic search, prediction‑bound drawers, belief ranking, consolidation with forgetting curve, and a portable signed format — features that alternatives like MemPalace or ad‑hoc CLAUDE.md files do not fully cover.

What are the runtime requirements?

Python 3.11+ and a SQLite database with sqlite‑vec support. No other external dependencies or services are required.

Where does the data live?

All memory data is stored locally in a SQLite database on the user’s hard drive. It is not sent to any remote server.

What transport does the MCP server use?

The server communicates over stdio (standard input/output) using the Model Context Protocol.

How do I install and remove it?

Install: pip install cairntir then cairntir setup. Uninstall: pip uninstall cairntir. The setup wizard handles all MCP registration; the server’s launcher is stable across venv moves and Python upgrades.