The gap: Claude Code doesn't know your pipeline, inbox, or meetings
Claude Code is Anthropic's agentic coding tool that runs from a terminal session, built for developers and technical operators, not a system of record. If you're the kind of RevOps engineer or technical founder who already runs your whole sales operation from a terminal, Claude Code still has no native memory of your pipeline, inbox, or meeting history, because that was never its job. Connecting Klipy's MCP server adds that as a second surface Claude Code can read and act on, alongside your codebase, in the same session.
There's no bundled, one-click Claude Code connector for Klipy. The real mechanism is an open MCP endpoint that any client can add, and Claude Code is one of many clients that can add it. No /compare page exists for Claude Code, because there's nothing to replace: Claude Code stays your coding agent, and Klipy stays the record of your pipeline, inbox, and meetings. Claude Code calls Klipy's tools in one direction only. Klipy does not reach into or read any other MCP server Claude Code might also be connected to.
How Claude Code connects to Klipy
Add Klipy's remote MCP server with one terminal command, then authenticate with OAuth. There's no app to install and no native connector to configure.
Run claude mcp add --transport http klipy https://api.klipy.ai/mcp from your terminal. By default the server is scoped to the current project; add --scope user to make it available across every project on the machine instead.
Then authenticate. Inside a Claude Code session, run /mcp, select klipy, and follow the browser OAuth login, or authenticate directly from the shell with claude mcp login klipy. Working over SSH or another environment without a local browser prints an authorization URL instead of opening one: open that URL locally, then paste the redirect URL back at the prompt (use ssh -t, or pass --no-browser to force this flow).
The first connection shows an OAuth consent screen with a checkbox for each of Klipy's nine capability scopes: crm:read, crm:write, comms:read, comms:draft, messages:send, meetings:read, agent:read, agent:write, and account:read. Claude Code can only call the tools whose scopes are checked. One detail worth knowing: the active grant belongs to your Klipy account identity, not to Claude Code specifically, so it governs every MCP client connected to that identity. If you later connect a different agent and grant it broader scopes, Claude Code picks up that same broader access the next time it calls Klipy, since Klipy tracks one active grant per identity, not one per client.
Within the scopes granted, two kinds of actions behave differently. Message sends are hard-gated by the system itself: the only path to a real outbound email or message is the two-step prepare_message_send / send_message_draft handshake, which requires a single-use confirmation token issued for that exact draft. Record writes aren't gated the same way. Once crm:write is active, tools like update_crm_record and delete_crm_record execute immediately, relying on Claude Code's own judgment rather than a system-enforced confirmation step.
To verify the connection, ask Claude Code "Who am I in Klipy?" It calls get_current_user and returns your Klipy account details if the connection is live.