AllSource Prime is AI-native memory over the Model Context Protocol. It works with any MCP-compliant client — the same hosted memory across every tool your agents touch. Pick yours below for a copy-paste setup.
Mint an API key and Prime syncs to your AllSource tenant. The same key works across every client — watch nodes appear live in the dashboard.
Mint an API keyNo account needed. Run cargo install allsource-prime and keep memory on disk at ~/.prime/memory.
Wire Prime into Claude Code with one `claude mcp add` command — no config file to edit.
The original Prime surface — paste one JSON block, or double-click the .dxt bundle for a no-terminal install.
Cursor reads stdio MCP servers from the same mcpServers shape as Claude Desktop.
Windsurf's Cascade reads MCP servers from an mcp_config.json with the standard mcpServers envelope.
VS Code's built-in MCP support (Copilot agent mode) loads stdio servers from a workspace .vscode/mcp.json.
Use Prime from ChatGPT desktop via its MCP connector — the same local stdio binary.
OpenAI's Codex CLI reads MCP servers from ~/.codex/config.toml under an [mcp_servers] table.
OpenCode uses its own envelope: a top-level `mcp` key, `type: "local"`, and command-as-array.
Using a client that isn't listed? Any MCP-compliant tool works — point it at the allsource-prime stdio binary. See the MCP setup docs for the generic config shape.