cursor.directory
BYTEROVER logo

BYTEROVER

Shared memory layer for your AI coding agents - integrated with all major AI IDEs --ByteRover helps create a shared memory layer that captures and indexes all agent interactions, reasoning paths, and developer feedback during coding tasks. Then from those experiences, we creates a memory layer stored in a vector database. Those memories could be retrieved as the AI agents encounter similar contexts to help them find the optimal solution tailored for that specific project/company requirements.

Installation Instructions

How to Use Model Context Protocol (MCP) in Cursor

What is MCP?

Model Context Protocol (MCP) is an open protocol that allows you to provide custom tools to agentic LLMs (Large Language Models) in Cursor's Composer feature.

Installation Steps

  1. Open Cursor Settings
    • Navigate to Cursor Settings > Features > MCP
    • Click the "+ Add New MCP Server" button
  2. Configure the Server
    • Name: Give your server a nickname
    • Type: Select the transport type (stdio or sse)
    • Command/URL: Enter either:
      • For SSE servers: The URL of the SSE endpoint
      • For stdio servers: A valid shell command to run the server

Example Configurations

For stdio Server (Weather Server Example):

Command: node ~/mcp-quickstart/weather-server-typescript/build/index.js

For SSE Server:

URL: http://example.com:8000/sse

Using MCP Tools

  1. Tool Availability
    • After adding a server, it will appear in your MCP servers list
    • You may need to click the refresh button to populate the tool list
  2. Using Tools in Composer
    • The Composer Agent automatically uses MCP tools when relevant
    • You can explicitly prompt tool usage by:
      • Referring to the tool by name
      • Describing the tool's function
  3. Tool Execution Process
    • Displays a message in chat requesting approval
    • Shows tool call arguments (expandable)
    • Executes the tool upon user approval
    • Displays the tool's response in the chat

Important Notes

  • MCP tools may not work with all models
  • MCP tools are only available to the Agent in Composer
  • For servers requiring environment variables, create a wrapper script that sets the variables before running the server