Uteke — Claude Code MCP server: memory and setup
Uteke — an MCP server for Claude Code that stores details from past work and retrieves them by meaning or keywords. The agent can query that memory to recall a saved fact.
Apache-2.0 ★ 291 updated 2026-11-16
claude mcp add --transport http uteke https://uteke.app/mcpWhat Uteke does
Uteke keeps information for later retrieval, matching requests by meaning as well as text. Its description also mentions rooms and time-travel through stored history.
For Uteke Cloud, connect the server with claude mcp add --transport http uteke https://uteke.app/mcp. The README also describes a local setup using the Uteke binary and the MCP command uteke-mcp; its first run downloads an embedding model of about 200 MB. The local option runs without an API key.
What it helps with
- retain details from earlier conversations for later use
- retrieve a note when the wording differs from the original
- look up a saved fact using a keyword
Example prompts for the agent
- Remember that the release moved to Friday.
- Find when we planned the release.
- Recall what was saved about the project setup.
How to install Uteke
Platform CLI: remote server
claude mcp add --transport http uteke https://uteke.app/mcpPlatform CLI: remote server
codex mcp add uteke --url https://uteke.app/mcpConfig file
{
"mcpServers": {
"uteke": {
"url": "https://uteke.app/mcp"
}
}
}Config file
{
"servers": {
"uteke": {
"type": "http",
"url": "https://uteke.app/mcp"
}
}
}Platform CLI: remote server
gemini mcp add -t http uteke https://uteke.app/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Uteke
How do I install Uteke in Claude Code?
To connect through Uteke Cloud, run claude mcp add --transport http uteke https://uteke.app/mcp.
What does Uteke do?
It gives an AI agent long-term memory and retrieves stored details through meaning-based and full-text search.
Does Uteke need an API key?
The local setup does not need one. Its first run downloads an embedding model of about 200 MB.