mcp-state

为智能体提供嵌入式SQLite黑板，使AI智能体能够读取、修补和归档其状态，而不会使对话上下文变得杂乱。

文档

mcp-state

mcp-state is a Model Context Protocol (MCP) server for agent state management. It provides a lightweight, deterministic "blackboard" architecture powered by an embedded SQLite database, enabling AI agents to read, patch, and archive their state without cluttering their conversation context.

Documentation

Full Diataxis documentation, interactive API playgrounds, and architectural guides are available at the mcp-state Documentation Site.

Features

Embedded SQLite Core: Zero database servers to manage. Everything is stored in a centralized blackboard.db file in your user data directory.
JSON Patching: Agents update state incrementally via JSON-RPC.
Cold Storage Archiving: Agents can archive completed subtasks to a cold_storage table to preserve their context window limits.
Native stdio MCP Server: Fully compliant with MCP 2024-11-05 standard over standard I/O streams.

Installation

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/JohnnytheShark/mcp-state/master/install.sh | bash

Windows PowerShell

irm https://raw.githubusercontent.com/JohnnytheShark/mcp-state/master/install.ps1 | iex

From Source

cargo build --release

Quick Start

mcp-state runs over stdio and is designed to be invoked by an MCP client (such as Claude Desktop or Antigravity).

{
  "mcpServers": {
    "mcp-state": {
      "command": "mcp-state",
      "args": []
    }
  }
}

The server exposes the following tools:

read_blackboard(session_id)
patch_blackboard(session_id, json_patch)
archive_subtask(session_id, key_to_archive)

License

This project is licensed under the Apache License 2.0. See the LICENSE and NOTICE files for details.