Continue.dev

February 25, 2026 ยท View on GitHub

Continue has native MCP support with multiple transport options. However, it likely does not inject MCP server instructions, so you need a one-line system prompt addition for proactive memory behavior.

MCP Setup

Add to your Continue configuration (.continue/config.json or via the UI):

{
  "mcpServers": [
    {
      "name": "mnemory",
      "transport": {
        "type": "streamable-http",
        "url": "http://localhost:8050/mcp",
        "headers": {
          "Authorization": "Bearer your-api-key",
          "X-Agent-Id": "continue"
        }
      }
    }
  ]
}

System Prompt

Continue likely does not inject MCP server instructions. Add this to your system prompt or model configuration:

Always call initialize_memory at the start of each conversation and follow received instructions for further memory interactions.

The initialize_memory tool returns behavioral instructions and core memories in one call, so the LLM knows how to use all memory tools effectively.

See system prompt templates for more detailed options.

How Memory Works

Once the system prompt line is added, Continue will:

  1. Load context at session start -- calls initialize_memory
  2. Search before answering -- when questions benefit from personal or project context
  3. Store new information -- architecture decisions, coding conventions, project structure

Tips

  • Cross-tool memory -- memories are shared across Continue, Claude Code, Cursor, and any other MCP client
  • Use project:<name> categories -- tag project-specific memories for easy filtering
  • Model flexibility -- Continue supports many LLM providers; mnemory works with all of them