Memava is not loyal to any one AI. Your vault speaks the open Model Context Protocol, so every tool below connects as a named team member with its own permissions, and they can all work the same vault at once. Each tool connects as its own agent; how many agents a vault holds depends on its plan, and the trial holds 10.
The always-on kind. An agent that runs your operations all day needs a memory that outlives the day, and this is the section it retires its amnesia in.
One protocol is the whole trick: we never build one-off plugins that rot, and you never wait for us to support your agent.
Every connection needs two things: your endpoint URL and an agent token. Both live inside your vault: sign in, hit Connect AI in the menu bar, and copy them (or the whole ready-made snippet for your tool). Every vault is born with its first agent already minted, every agent gets its own token, and revoking one never touches the others. The endpoint looks like:
https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN
Full page: /connect/claude
Settings, then Connectors, then "Add custom connector". Paste your endpoint URL as the remote MCP server. Claude can then read and write your vault in any chat where you enable it.
Full page: /connect/claude-code
claude mcp add --transport http memava \ https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN
Done. Your sessions can now list, read, search, and update notes inside the agent's grants.
Full page: /connect/codex
[mcp_servers.memava] url = "https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN"
Goes in ~/.codex/config.toml. The Codex CLI, its IDE extension and the ChatGPT desktop app all read that file, so one block connects all three.
Full page: /connect/chatgpt
On plans that allow custom connectors: Settings, then Connectors, then add a custom MCP connector with your endpoint URL. Availability depends on your ChatGPT plan and workspace policy.
Full page: /connect/cursor
{
"mcpServers": {
"memava": {
"url": "https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN"
}
}
}
Goes in .cursor/mcp.json in your project, or the global one in your home directory.
Full page: /connect/vscode
{
"servers": {
"memava": {
"type": "http",
"url": "https://memava.com/o/YOUR-TEAM/mcp/YOUR-AGENT-TOKEN"
}
}
}
Goes in .vscode/mcp.json. Works with Copilot's agent mode and MCP-aware extensions.
Full pages: Windsurf, Zed, Cline, Continue, Gemini CLI
Each has an MCP servers section in its settings; add a remote (streamable HTTP) server with your endpoint URL. The names differ, the shape is identical: one URL, no local process to babysit.
Every MCP-capable harness takes the same single line: add a remote (streamable HTTP) MCP server with your endpoint URL, in whatever its settings call that section. Your agent then reads and writes the vault as a named member, inside its grants.
Any MCP client library works: initialize against the endpoint, list tools, call them. Twenty-three tools, plain JSON-RPC over HTTPS. The Developers page has the full tool list and the trust model.
Stuck on any of these? Every setup page has the exact steps, and the contact form reaches us if a client behaves differently.
No. It holds what your people and agents write into the vault; nothing reads your pushes or pull requests. To keep that record, have your coding agent end each session with a short log of what it did and what is left, and every other tool can read it.
Yes. Connect each one as its own agent with its own token. What one writes, the others read on their next call, and the history names which tool wrote each line.
Ask it. The activity tool returns every recent change on notes it may see, with the author and the time, so a question like what did the team change this week is answered from the record.
Anything that speaks MCP: Claude, Claude Code, ChatGPT where custom connectors are switched on, Codex, Cursor, VS Code, Windsurf, Zed, Cline, Continue, Gemini CLI, and your own agents.