Install guide and config at curatedmcp.com

threadctx-mcp: Give Your AI Coding Agent Project Memory

Cursor and Claude Code are powerful, but they forget context between sessions. You end up re-explaining the same architectural decisions, coding conventions, and project quirks every time you start a new conversation. threadctx-mcp fixes that by giving your AI agent a persistent memory system—one that works identically in Cursor, Claude Code, and other AI coding tools.

What It Does

threadctx-mcp is an MCP (Model Context Protocol) server that lets your AI agent store and retrieve project-specific knowledge through two simple tools: memory_write and memory_query. The agent calls these automatically during coding sessions to record learnings—naming patterns you prefer, architectural decisions, common gotchas in your codebase—and recall them when relevant.

By default, it runs locally with zero network calls (except to your LLM provider) and requires no signup. Your memory stays on your machine. Flip a switch for the Team plan, and you get semantic search across team memory—meaning your entire dev team benefits from collective project knowledge, not just keyword matching.

This unlocks a fundamentally different workflow: instead of babysitting an AI agent that treats each conversation as a blank slate, you're building institutional memory that gets smarter as your project evolves.

How to Install

Install via npm:

npx threadctx-mcp

Add to your Claude Desktop config (usually ~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

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

Cursor and Windsurf have their own config formats; see the full docs for your tool.

Real-World Use Cases

• Onboard new features faster: Your agent remembers that your project uses a specific folder structure, naming convention, and error-handling pattern. New feature requests get implemented in hours instead of days of back-and-forth.

• Consistent refactoring: Store architectural decisions ("we use dependency injection in services," "tests live in __tests__ folders") once, recall them across dozens of conversations. No more inconsistent code.

• Team alignment without meetings: Store common blockers, setup gotchas, and library quirks that usually live in Slack or wiki pages. Your whole team's AI agents reference the same knowledge base.

---

Full install guides for Claude Desktop, Cursor, Windsurf, and more at CuratedMCP.