Gnosem MCP server
{
"mcpServers": {
"gnosem": {
"type": "http",
"url": "https://gnosem.dev/mcp"
}
}
}{
"mcpServers": {
"gnosem": {
"type": "http",
"url": "https://gnosem.dev/mcp"
}
}
}{
"mcpServers": {
"gnosem": {
"type": "http",
"url": "https://gnosem.dev/mcp"
}
}
}{
"servers": {
"gnosem": {
"type": "http",
"url": "https://gnosem.dev/mcp"
}
}
}{
"mcpServers": {
"gnosem": {
"type": "http",
"url": "https://gnosem.dev/mcp"
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsonversion 1.0.0
What it does
Gnosem is a hosted MCP server that provides a single, persistent semantic memory store accessible across multiple AI clients and vendors (Claude, ChatGPT, Gemini, Cursor, etc.). It solves the problem of fragmented proprietary memory systems by allowing facts and preferences saved from one AI model to be read and written by any other MCP-capable client using the same API key.
- Save coding preferences, project decisions, or technical knowledge once and access them consistently across Claude, ChatGPT, Cursor, and other AI tools in the same session or future sessions.
- Maintain a shared context of user preferences and project decisions that persists across different AI vendors without manual re-entry of information.
- Search semantic memories across all saved facts to retrieve relevant context, decisions, or preferences based on natural language queries rather than exact text matching.
Tools this server exposes
| Tool | Does |
|---|---|
| memory_write | Save a fact, preference, decision, or note with optional tags and provenance tracking. |
| memory_search | Perform semantic search across saved memories using natural language queries. |
| memory_list | List recent memories in reverse-chronological order with optional pagination. |
| memory_forget | Soft-delete a memory by ID. |
| memory_supersede | Replace a stale memory with a corrected version while maintaining the correction chain. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.