Persistent Memory for Cursor AI with sovseal MCP

sovseal gives Cursor’s AI agent a persistent memory layer that outlives individual sessions. Once configured, the agent can recall your preferred frameworks, database choices, package version overrides, and design decisions without you having to re-explain them every time you open a project. Cursor supports MCP servers natively, so no signup or cloud configuration is required for the local memory tier.

Installation

You can add the sovseal MCP server through the Cursor UI or by editing the config file directly.

Cursor Settings UI
Config file

Open Cursor Settings

Click the gear icon in the top-right corner of the Cursor window.

Navigate to MCP

Go to Features → MCP, then click + Add New MCP Server.

Fill in the server details

Enter the following values:

Field	Value
Name	`sovseal-memory`
Type	`stdio`
Command	`npx -y @sovseal/mcp-server`

Save and reload

Click Save. Cursor will start the MCP server in the background. The store_memory and recall_memory tools become available in the Agent chat panel.

Add the following block to your global ~/.cursor/mcp.json for memory shared across all projects, or to .cursor/mcp.json inside a specific project root for project-scoped memory:

~/.cursor/mcp.json

{
  "mcpServers": {
    "sovseal-memory": {
      "command": "npx",
      "args": ["-y", "@sovseal/mcp-server"]
    }
  }
}

Save the file and restart Cursor to apply the change.

Add proactive recall with Cursor Rules

By default, the agent only calls memory tools when you explicitly ask. To make Cursor automatically recall relevant context before every response, add the following instructions to Cursor Settings → General → Rules for AI, or create a .cursorrules file at the root of your project:

.cursorrules

Always invoke the `recall_memory` tool as your first action on any new user request
to retrieve historical context, past code decisions, and user stack preferences.
Do not ask the user for context if it can be retrieved via a recall query.
When the user shares a persistent stack choice, API change, or file path decision,
call `store_memory` immediately to persist it for future sessions.

Project-scoped vs. global memory

By default, sovseal writes all memories to ~/.sovseal/db/memories.lance, which is shared across every project. If you want to prevent React Native context from bleeding into a Rust project, set the SOVSEAL_DB_DIR environment variable to a project-local path:

.cursor/mcp.json

{
  "mcpServers": {
    "sovseal-memory": {
      "command": "npx",
      "args": ["-y", "@sovseal/mcp-server"],
      "env": {
        "SOVSEAL_DB_DIR": "./.sovseal/db"
      }
    }
  }
}

Memories stored with this config are written to .sovseal/db/ inside the project folder and are never visible to other projects.

Troubleshooting

Tool fails to load in Cursor Chat

Node.js version: Cursor spawns MCP servers from its own background process using your system’s global Node.js. Run node -v in a terminal and confirm it reports >= 20.0.0.
Warmup latency: On the first run, the ONNX embedding model (~22 MB) downloads to ~/.sovseal/models/. The tool status in Cursor may show “Connecting…” for up to 30 seconds during this download — this is expected and happens only once.
Inspect the logs: Open Cursor’s developer console via Help → Toggle Developer Tools → Console and look for any MCP server error output.

Get Started

Platform

Agent Plugins

Self-Hosted

Integrations

Cookbooks

API Reference

Release Notes

Persistent Memory for Cursor AI with sovseal MCP

Installation

Add proactive recall with Cursor Rules

Project-scoped vs. global memory

Troubleshooting

​Installation

​Add proactive recall with Cursor Rules

​Project-scoped vs. global memory

​Troubleshooting

Installation

Add proactive recall with Cursor Rules

Project-scoped vs. global memory

Troubleshooting