1echa/context-sync-mcp

Developer tools

This page was translated automatically. Technical details may differ from the original. View the original

20 stars 3 forks Quality 69 Trend 10MIT

Repository Install

Cross-device context sync for coding agents via MCP + Git |

Overview

Never Repeat Yourself to AI Again. (Give your Codex / Claude Code / Gemini / Antigravity long-term memory that never forgets across devices, agents, and devices) ❌ : In the morning, I used Claude Code, and after much effort, the agent finally figured out the logic of the codebase and knew which internal API would cause errors; But if you switched computers or models too late, your Coding Agent instantly lost its memory. Then you had to enter it again in a new dialog box, only to find the model still couldn't remember what pitfalls the agent had before or what wonderful development time you had. :d isguised_face:, then it started telling it what had been done before, which pitfalls you shouldn't fall into after the mistakes, leaving you physically and mentally exhausted. ✅ :Context Sync provides a ** for the Agent. Based on Git and MCP, switching computers and agents, typing /sync-load can keep progress and context connected. For a new project, first sync_init establish the first context → hit a pitfall or make a new decision, use write_context to automatically record → sync_push push → switch computers sync_load pull it. It's that simple. It doesn't synchronize "private chat records of a particular agent," but rather the structured context files within the project; So the same context can be reused between . 💡 Hands-on demonstration: from Antigravity-Claude Code Opus 4.6 to GPT-5.4 - : Execute /sync-init - : Have the Agent call write_context record to .context/ - : Execute /sync-save - : Execute /sync-load : A full day of daily development (5-10 writes + 1 synchronization) consumes only about 2k-5k tokens in total, with almost no impact on the bill. Keep up with the times. By default, the prompt will include a supplementary prompt. If you need to write globally in the project prompt, you can modify the global prompt - ≥ 18.0.0 - SSH/HTTPS authentication installed and configured - The current directory is in a Git repository

Overview (machine translated)

🔄 Context Sync MCP

Never Repeat Yourself to AI Again.
(Give your Codex / Claude Code / Gemini / Antigravity long-term memory that never forgets across devices, agents, and devices)

🎯 Why is it needed?

❌ > Pain Points: In the morning, I used Claude Code, and after much effort, the agent finally figured out the logic of the codebase and knew which internal API would cause errors; But if you switched computers or changed models later, your coding agent instantly lost memory. Then you had to enter it again in a new dialog box, but found the model still couldn’t remember what pitfalls the agent had before or what kind of wonderful development time it had with you. :d isguised_face:, then it started telling it what had been done before, which pitfalls you shouldn’t tackle, and you were completely exhausted.

✅ > Solution: Context Sync provides Agents with a long-term memory library independent of conversation and specific models. Based on Git and MCP, switching computers and agents, typing /sync-load, can continue progress and context.

Minimalist Workflow: For new projects, first sync_init establish the first context → Hit a pitfall / New decisions use write_context to automatically record → sync_push push → switch computers sync_load pull data. It’s that simple.

It syncs not “private chat records of a particular agent,” but structured context files within the project; So the same context can be reused between Claude ↔ Codex ↔ Gemini.

💡 Hands-on demonstration: from Antigravity-Claude Code Opus 4.6 to GPT-5.4

🚀 Recommended Workflow

First Connection of New Projects: Execute /sync-init
New context added in daily development: Let the Agent call write_context to record .context/
Save before leaving the current device: execute /sync-save
Restoring on another device: Execute /sync-load

Simple memorization: First init, keep writing, before switching, save, new device load.

✨ Core Features

Features	Description
🧠 Structured Long-Term Memory	Distill scattered conversations into five types of standard documents (such as warning points, architectural decisions, etc.), and refuse to forget.
🔗 Association and Automatic Numbering	Pitfall records automatically increment numbering, supporting `related_to` bidirectional jumping, building the project’s knowledge graph.

This page was translated automatically. Technical details may differ from the original.

Install

This server does not publish a one-line install command.

Open the repository installation guide New to MCP? Start with the mcpskill.org guide →

Configuration

{
  "mcpServers": {
    "context-sync": {
      "command": "node",
      "args": ["/绝对路径/context-sync-mcp/dist/index.js"]
    }
  }
}