Skip to main content
Curtly provides an official Model Context Protocol (MCP) server, allowing AI coding assistants and agent environments to automatically compress workspace context, files, and multi-turn prompt buffers before making inference calls.

1-Click Client Configurations

Cursor (.cursor/mcp.json)

Add the following to your Cursor MCP settings (~/.cursor/mcp.json or project-level .cursor/mcp.json):

Claude Desktop / Claude Code (claude_desktop_config.json)

Add Curtly to your Claude configuration file:

Windsurf & OpenCode


Available MCP Tools

When connected, the Curtly MCP server exposes the following tools to the LLM agent:

1. curtly_compress

Compresses any arbitrary text prompt or context block with deterministic AST Safe Vault code protection. Parameters:
  • prompt (string, required): The raw prompt text to compress.
  • mode (string, optional): 'conservative' | 'balanced' (default) | 'aggressive'.
  • protectCodeBlocks (boolean, optional, default: true): Locks code blocks into the Safe Vault.
  • protectJson (boolean, optional, default: true): Locks JSON schemas into the Safe Vault.

2. curtly_get_metrics

Returns real-time session telemetry including total input tokens saved, estimated output tokens saved, and net cost savings in USD.