Memory Crystal Demo
connect
two tools. any model. 15 minutes of memory.
your memory. your machine. every model.
MCP server URL
https://memory-crystal-demo.wipcomputer.workers.dev/mcp
demo memories expire in 15 minutes. the real version doesn't.
Claude
claude code cli
1. run this in your terminal:
claude mcp add --transport http crystal-demo https://memory-crystal-demo.wipcomputer.workers.dev/mcp
2. start a conversation. you'll see weave and pull in your tools.
claude.ai web + desktop
1. go to settings → connectors → add custom connector
2. paste the MCP server URL above
3. start a new conversation. weave and pull are available.
claude ios + android
1. add the connector on claude.ai (web) using the steps above
2. it syncs to mobile automatically. no separate setup needed.
OpenClaw
openclaw agent runtime
1. add as an MCP extension in openclaw.json or as a plugin
"mcp": { "servers": { "crystal-demo": { "url": "https://memory-crystal-demo.wipcomputer.workers.dev/mcp" } } }
2. restart the gateway: openclaw gateway restart
3. your agent now has weave and pull. works with any model backend.
OpenClaw is a sovereign agent runtime. Memory Crystal was built on it. github.com/openclaw/openclaw
ChatGPT
chatgpt with mcp
1. go to settings → tools → add MCP server
2. paste the MCP server URL
3. start a conversation. weave and pull are available as tools.
MCP support in ChatGPT is rolling out. If you don't see the option yet, check OpenAI's docs for the latest.
Grok
grok via mcp client
1. use any MCP-compatible TUI or client that supports Grok as a backend
2. add the MCP server URL to your client config
3. start a session. weave and pull work like any other MCP tool.
Grok doesn't natively support MCP connectors yet, but any TUI that routes through Grok (like Claude Code alternatives or custom setups) can connect.
Gemini
gemini via mcp client
1. use any MCP-compatible client that supports Gemini as a backend
2. add the MCP server URL to your client config
3. start a session. weave and pull are standard MCP tools.
Google is adding MCP support to Gemini. Check their docs for the latest connector setup.
any model
any mcp-compatible client
1. the demo server speaks standard MCP over Streamable HTTP
2. point any MCP client at the URL. no auth. no tokens.
3. you get two tools: weave and pull. that's it.
if your client speaks MCP, it works. model doesn't matter. that's the point.