Install Ricord — pick your client
Three commands and a smoke test, every time. Pick your MCP-compatible client below for the per-client config paths, gotchas, and verification steps.
The fastest path is two commands:
npm install -g ricord ricord setup # auto-detects every MCP client on your machine
ricord setup logs you in, then writes the right config block in the right place for every MCP client it finds. The per-client guides exist for the verification steps, the smoke tests, and the gotchas — particularly where PATH inheritance, sandbox profiles, or per-OS config paths differ.
Per-client guides
Install in Claude Desktop
Per-OS config paths (Mac / Win / Linux), MCP hammer-icon verification, 30-second smoke test.
Install in Cursor
Global vs per-project mcp.json, the Cursor PATH gotcha, Composer smoke test.
Install in Claude Code
Canonical claude mcp add path, user-scope vs project-scope, in-terminal smoke test.
Install in Codex CLI
TOML config (not JSON), working inside the sandbox, replacing improvised CONTEXT.md.
Install in Zed
context_servers in settings.json, AI-panel smoke test, the Zed-from-Finder PATH gotcha.
Install in Gemini CLI
mcpServers in ~/.gemini/settings.json, /mcp verification, workspace vs user scope.
Install in Windsurf
mcpServers in ~/.codeium/windsurf/mcp_config.json, Cascade refresh, Chat vs Write mode notes.
Install in Cline
OSS VS Code agent (formerly Claude Dev). MCP Servers panel + autoApprove tuning + VS-Code-from-Finder PATH gotcha.
Don't see your client?
Any client that speaks the Model Context Protocol can connect to Ricord. The MCP server config looks the same everywhere — a command, an args array, optional env. The block to add is:
{
"mcpServers": {
"ricord": {
"command": "ricord",
"args": ["mcp"]
}
}
}Or the TOML equivalent for clients that use TOML (Codex): [mcp_servers.ricord] with command = "ricord" and args = ["mcp"]. If your client supports MCP but isn't auto-detected by ricord setup, drop a note in the docs and we'll add it.
Cross-client memory
The same ricord setup wires every MCP client on your machine to the same memory backend. Switch between Claude Desktop and Cursor mid-debugging, or jump from a Codex session in the terminal to a Composer chat in Cursor — the wiki view and recall are consistent across all of them.
Memory is scoped per project automatically, so the four side-projects you work on in different tabs don't leak into each other.