> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs-beta.getzep.com/v3/changelog/2026/7/31/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs-beta.getzep.com/_mcp/server. ## Connect every project through one Memory MCP endpoint Memory MCP clients now use one universal endpoint — `https://api.getzep.com/mcp` on managed cloud, or `/mcp` on your BYOC API host — instead of a per-project URL. Zep asks for the user's work email, discovers the identity provider from that domain, and offers an account and project chooser when more than one project is eligible. Email-domain routing for Custom OIDC uses domains assigned to the account's identity provider configuration. A connection's allowed domains list only narrows which of those domains the project admits. Google Workspace connections use the hosted domain verified during administrator setup. Connections also expose a grant maximum lifetime (default 24 hours) so an actively refreshing client must re-authenticate against the live identity provider when the ceiling elapses. Read the docs ## Find standalone graphs by name and description Standalone graphs now support names and descriptions as routing metadata. The `graph.list_all` operation searches `graph_id`, `name`, and `description` and returns deterministic pagination metadata. Graph listing now returns 50 results when you omit `pageSize`, instead of the previous unbounded result. Set `pageNumber` and `pageSize` to retrieve additional pages. Valid `pageSize` values are 1 through 100; explicitly passing `0` now returns a validation error instead of an unbounded result. Search queries longer than 200 Unicode code points after whitespace normalization now return a validation error instead of being silently truncated. When standalone graph access is enabled, MCP clients can read `zep://graphs/directory` or call `list_graphs`, then pass the returned `graph_id` to `search_graph_in`. Read the docs